Reversibility, undoing transformations, the inverse formula, row reduction methods, singular matrices, and the connection between invertibility and unique solutions
This chapter develops one of the most powerful ideas in linear algebra: the notion that some operations can be undone. When you multiply a number by 3, you can undo that operation by multiplying by one-third. When a function applies a transformation, an inverse function reverses it. Matrices follow the same pattern. If a matrix represents a transformation or encodes a system, an inverse matrix — when it exists — reverses that transformation and solves that system.
Chapter 5 builds directly on the matrix operations of Chapter 4 and the system-solving techniques of Chapter 3. It connects the algebraic machinery of matrices to a deeper structural idea: reversibility. Not every matrix has an inverse, and understanding why is just as important as knowing how to find one when it does exist.
By the end of this chapter, students will be able to find inverse matrices, use them to solve linear systems, and recognize the conditions under which a matrix fails to be invertible.
Suppose a company uses a matrix equation to model how resources are allocated across two departments. The matrix equation is Ax = b, where A is a known 2 × 2 matrix and b is a known vector representing target output. The company wants to find the input vector x. If the matrix A were a number, the answer would be easy: just divide both sides by A. But matrices cannot be divided. So what do we do? The answer lies in the inverse matrix.
AA⁻¹ = A⁻¹A = I.Ax = b using an inverse matrix.Inverse matrix · Invertible matrix · Singular matrix · Noninvertible matrix · Identity matrix · Matrix equation · Elementary matrix · Row reduction (inverse method) · Augmented matrix [A | I] · Unique solution · Reversible transformation
To understand the inverse of a matrix, begin with what you already know about inverses in arithmetic and algebra. Every nonzero number a has a multiplicative inverse, the number 1/a, also written a⁻¹. The defining property is:
The number 1 is the multiplicative identity — it does nothing when multiplied. Multiplying by a⁻¹ undoes the effect of multiplying by a. Matrices have an analogous structure. For square matrices, the identity matrix plays the role of the number 1.
An n × n identity matrix, written Iₙ or simply I, has ones on the main diagonal and zeros everywhere else. For any compatible matrix A: AI = IA = A. The identity matrix does nothing.
2 × 2: I₂ = [[1, 0], [0, 1]]
3 × 3: I₃ = [[1, 0, 0], [0, 1, 0], [0, 0, 1]]
Let A be a square n × n matrix. A matrix B is called the inverse of A if:
AB = Iₙ and BA = Iₙ
When such a matrix B exists, we write B = A⁻¹ and say that A is invertible. The notation A⁻¹ does not mean 1/A — matrices cannot be divided. It means the specific matrix that produces the identity when multiplied on either side of A.
Both conditions — AA⁻¹ = I and A⁻¹A = I — are required because matrix multiplication is not commutative. For square matrices, if AB = I then BA = I automatically holds, but this symmetry fails for non-square matrices, which is why inverses are defined only for square matrices.
If a matrix A represents a geometric transformation — a rotation, reflection, scaling, or shear — then A⁻¹ represents the transformation that undoes it.
| Transformation A Performs | What A⁻¹ Does |
|---|---|
| Rotates every vector 60° counterclockwise | Rotates every vector 60° clockwise |
| Stretches horizontally by a factor of 3 | Compresses horizontally by a factor of 1/3 |
| Reflects across the line y = x | Same reflection — it undoes itself |
A matrix A can be thought of as a machine. You feed in a vector x, and the machine produces Ax. The inverse matrix A⁻¹ is the reverse machine: you feed in Ax, and it produces x again.
This undoing property is exactly what allows us to solve matrix equations. If Ax = b, and we know A⁻¹, then multiplying both sides on the left by A⁻¹ gives:
If A and B are both invertible n × n matrices:
1. (A⁻¹)⁻¹ = A — the inverse of A⁻¹ is A itself
2. (AB)⁻¹ = B⁻¹A⁻¹ — the inverse of a product reverses order
3. (Aᵀ)⁻¹ = (A⁻¹)ᵀ — the inverse of the transpose is the transpose of the inverse
4. (cA)⁻¹ = (1/c)A⁻¹ for nonzero scalar c
Property 2 — that (AB)⁻¹ = B⁻¹A⁻¹ — has a memorable everyday analogy. Putting on socks, then shoes, must be undone by first removing shoes and then removing socks. To reverse a composed action, you reverse the last step first, then the first step. The reversal of order is required.
Not every matrix has an inverse. This is a fundamental fact of matrix algebra, and understanding it will clarify much of what follows in this chapter and the rest of the textbook.
In arithmetic, every real number has a multiplicative inverse except zero. For matrices, the situation is more complex. There are many matrices that, like the number zero, fail to have inverses. These are called singular matrices or noninvertible matrices.
A square matrix A is invertible (also called nonsingular) if there exists a matrix A⁻¹ such that AA⁻¹ = A⁻¹A = I. Otherwise, A is singular.
Geometrically, a matrix is singular if its associated transformation collapses space — if it maps two or more distinct vectors to the same output vector, making the transformation impossible to reverse. Consider a matrix that sends every vector to a single line. No matter what the input was, the output lies on that line. There is no way to look at the output and recover the original input, because many different inputs all produce the same output. You cannot reverse a many-to-one process.
Algebraically, a matrix is singular when its rows (or columns) are linearly dependent — when one row is a scalar multiple of another, or a combination of the others. This means the matrix contains redundant information and does not span enough directions to encode a unique solution.
| Matrix | Singular or Invertible? | Reason |
|---|---|---|
[[2, 4], [1, 2]] | Singular | Row 2 is exactly half of row 1 — linearly dependent rows |
[[3, 1], [2, 5]] | Invertible | Rows point in genuinely different directions — independent |
[[1, 0], [0, 0]] | Singular | Collapses every vector onto the x-axis — loses the y component |
For a 2 × 2 matrix, there is a direct formula for the inverse. The formula requires only the four entries of the matrix and one important number — the determinant.
Let A = [[a, b], [c, d]]. If ad − bc ≠ 0, then A is invertible and:
A⁻¹ = (1 / (ad − bc)) · [[d, −b], [−c, a]]
The quantity ad − bc is the determinant of A, written det(A). If det(A) = 0, then A is singular and has no inverse.
The procedure: swap the main diagonal entries (a and d), negate the off-diagonal entries (b and c), and divide the entire matrix by the determinant.
Step 1 — Compute the determinant:
det(A) = (3)(5) − (1)(2) = 15 − 2 = 13
Since 13 ≠ 0, the matrix is invertible.
Step 2 — Apply the inverse formula:
A⁻¹ = (1/13) · [[5, −1], [−2, 3]]
= [[ 5/13, −1/13],
[−2/13, 3/13]]
Row 1, Col 1: (3)(5/13) + (1)(−2/13) = 15/13 − 2/13 = 13/13 = 1 ✓
Row 1, Col 2: (3)(−1/13) + (1)(3/13) = −3/13 + 3/13 = 0 ✓
Row 2, Col 1: (2)(5/13) + (5)(−2/13) = 10/13 − 10/13 = 0 ✓
Row 2, Col 2: (2)(−1/13) + (5)(3/13) = −2/13 + 15/13 = 13/13 = 1 ✓
det(B) = (4)(3) − (6)(2) = 12 − 12 = 0
A⁻¹ mean, and what does it not mean?Inverse matrices provide a powerful method for solving matrix equations of the form AX = B, where A, X, and B are matrices of compatible sizes and A is invertible.
Solving AX = B: Multiply both sides on the left by A⁻¹:
A⁻¹(AX) = A⁻¹B → IX = A⁻¹B → X = A⁻¹B
Solving XA = B: Multiply both sides on the right by A⁻¹:
XA · A⁻¹ = B · A⁻¹ → X = BA⁻¹
Because matrix multiplication is not commutative, the side on which you multiply by A⁻¹ matters. For AX = B, the solution is X = A⁻¹B (left-multiply). For XA = B, the solution is X = BA⁻¹ (right-multiply). These give different results. Never switch the order without justification.
X = A⁻¹B = (1/13) · [[5, −1], [−2, 3]] · [[7], [8]]
= (1/13) · [[(5)(7) + (−1)(8)], [(−2)(7) + (3)(8)]]
= (1/13) · [[35 − 8], [−14 + 24]]
= (1/13) · [[27], [10]]
= [[27/13], [10/13]]
[[3, 1], [2, 5]] · [[27/13], [10/13]]
= [[81/13 + 10/13], [54/13 + 50/13]]
= [[91/13], [104/13]]
= [[7], [8]] ✓
The most important application of matrix inverses in this chapter is solving the linear system Ax = b, where A is a known n × n matrix, b is a known vector, and x is the unknown vector. If A is invertible, the unique solution is:
This is a direct, explicit formula for the solution. Once you have A⁻¹, solving the system for any right-hand side b becomes a single matrix-vector multiplication.
| When the Inverse Method Is Useful | When Row Reduction Is Preferred |
|---|---|
| Solving Ax = b for many different vectors b with the same A | Solving the system only once |
| When A⁻¹ is already known or easy to compute | When A is large (fewer operations) |
| When you want an explicit formula for the solution | When numerical precision is important |
x = A⁻¹b = (1/13) · [[5, −1], [−2, 3]] · [[10], [4]]
= (1/13) · [[(5)(10) + (−1)(4)], [(−2)(10) + (3)(4)]]
= (1/13) · [[50 − 4], [−20 + 12]]
= (1/13) · [[46], [−8]]
= [[46/13], [−8/13]]
Eq. 1: 3(46/13) + 1(−8/13) = 138/13 − 8/13 = 130/13 = 10 ✓
Eq. 2: 2(46/13) + 5(−8/13) = 92/13 − 40/13 = 52/13 = 4 ✓
Write each system as Ax = b. Find A⁻¹ and use it to solve. Then solve again for the second right-hand side provided.
For matrices larger than 2 × 2, the direct formula approach is not practical. Instead, we use a systematic method based on row reduction. The key idea is to track two processes simultaneously: reducing A to the identity while applying the same operations to an identity matrix beside it.
Step 1. Form the augmented matrix [A | Iₙ], appending the identity matrix to the right of A.
Step 2. Apply elementary row operations to the entire augmented matrix.
Step 3. The goal is to reduce A (the left side) to the identity matrix Iₙ.
Step 4. When this is achieved, the right side will contain A⁻¹:
[A | I] →(row operations)→ [I | A⁻¹]
If the left side cannot be reduced to I — if it produces a row of zeros — then A is singular and has no inverse.
Starting augmented matrix [A | I]:
[ 1 2 0 | 1 0 0 ]
[ 0 1 3 | 0 1 0 ]
[ 0 0 1 | 0 0 1 ]
R₂ ← R₂ − 3R₃:
[ 1 2 0 | 1 0 0 ]
[ 0 1 0 | 0 1 −3 ]
[ 0 0 1 | 0 0 1 ]
R₁ ← R₁ − 2R₂:
[ 1 0 0 | 1 −2 6 ]
[ 0 1 0 | 0 1 −3 ]
[ 0 0 1 | 0 0 1 ]
Left side = I, so right side = A⁻¹.
A · [[1], [0], [0]] = [[1], [0], [0]] ✓
A · [[-2], [1], [0]] = [[-2+2], [1+0], [0]] = [[0], [1], [0]] ✓
[ 1 2 3 | 1 0 0 ]
[ 2 4 6 | 0 1 0 ]
[ 0 1 1 | 0 0 1 ]
R₂ ← R₂ − 2R₁:
[ 1 2 3 | 1 0 0 ]
[ 0 0 0 | −2 1 0 ]
[ 0 1 1 | 0 0 1 ]
Row 2 is all zeros on the left side.
The left portion cannot be reduced to I.
Every row operation applied during row reduction can be represented as multiplication by a special matrix called an elementary matrix. There are three types, each corresponding to one of the three allowed row operations.
| Row Operation | Elementary Matrix Construction |
|---|---|
| Scale a row by nonzero constant c | Identity matrix with one diagonal entry replaced by c |
| Swap two rows | Identity matrix with two rows exchanged |
| Add a multiple of one row to another | Identity matrix with one off-diagonal entry set to c |
If E is an elementary matrix, then EA is the result of applying the corresponding row operation to A. Elementary matrices are always invertible. Their inverses are also elementary matrices corresponding to the reverse operation.
The row reduction sequence can be written as: Eₖ···E₂E₁A = I, which means Eₖ···E₂E₁ = A⁻¹. The right side of the augmented matrix tracks this product automatically.
I₃ = [[1, 0, 0], E = [[1, 0, 0],
[0, 1, 0], [2, 1, 0],
[0, 0, 1]] [0, 0, 1]]
A singular matrix is a square matrix that has no inverse. We have seen singular matrices arise when rows are linearly dependent, when row reduction produces a row of zeros, and when the determinant is zero. Here we gather the key conditions together.
A square matrix A is singular if any of the following holds:
1. Its determinant equals zero: det(A) = 0
2. Its rows are linearly dependent (one row is a combination of others)
3. Its columns are linearly dependent
4. Row reduction of [A | I] produces a row of zeros on the left
5. The equation Ax = 0 has a nonzero solution
6. The equation Ax = b either has no solution or infinitely many solutions for some b
These conditions are all equivalent. Chapter 10 will explain the structure that links them.
Imagine a transformation that takes a two-dimensional plane and maps it onto a single line. Every point on the plane gets sent to a point on that line. If you look at any output point and ask "where did this come from?", the answer is an entire line of input points — infinitely many points all map to the same output. You cannot reverse this process. Singular matrices are exactly the matrices that perform such a collapse: in two dimensions, they collapse a plane to a line or a point; in three dimensions, a 3D space to a plane, a line, or a point.
There is a deep connection between the invertibility of a matrix and the nature of solutions to the system Ax = b. The following theorem ties together many ideas and will grow as we develop more tools.
Let A be an n × n matrix. The following statements are equivalent — they are all true at the same time or all false at the same time:
1. A is invertible.
2. The equation Ax = b has a unique solution for every b ∈ ℝⁿ.
3. The equation Ax = 0 has only the trivial solution x = 0.
4. The reduced row echelon form of A is Iₙ.
5. det(A) ≠ 0.
This theorem will expand as we study rank, null space, span, and independence in Chapters 9–10.
det(A) = (2)(2) − (−1)(−4) = 4 − 4 = 0
A is singular.
Augmented matrix: [ 2 −1 | 1 ]
[−4 2 | −2]
R₂ ← R₂ + 2R₁: [ 2 −1 | 1 ]
[ 0 0 | 0 ]
Row of zeros on left — infinitely many solutions.
With free variable x₂ = t: x₁ = (1 + t)/2
Augmented matrix: [ 2 −1 | 1 ]
[−4 2 | 3 ]
R₂ ← R₂ + 2R₁: [ 2 −1 | 1 ]
[ 0 0 | 5 ]
Last row reads 0 = 5 — impossible. No solution.
Matrix inverses appear naturally whenever a linear system must be solved repeatedly for different right-hand sides. The following example shows how the inverse method simplifies repeated computation in applied contexts.
In a simple economic model, suppose an economy has two sectors and the production of each requires inputs from both. Let A encode how much each sector requires to produce one unit of output. Given a vector d of final demand, the production vector x satisfies (I − A)x = d. If (I − A) is invertible, the solution is x = (I − A)⁻¹d. The matrix (I − A)⁻¹ is computed once, and production for any demand vector follows by a single multiplication.
Step 1 — Find A⁻¹:
det(A) = (5)(4) − (2)(3) = 20 − 6 = 14
A⁻¹ = (1/14) · [[4, −2], [−3, 5]]
Step 2 — Solve for b₁ = [[120], [96]]:
x₁ = A⁻¹b₁ = (1/14) · [[4·120 − 2·96], [−3·120 + 5·96]]
= (1/14) · [[480 − 192], [−360 + 480]]
= (1/14) · [[288], [120]]
≈ [[20.6], [8.6]]
Step 3 — Solve for b₂ = [[200], [160]]:
x₂ = A⁻¹b₂ = (1/14) · [[4·200 − 2·160], [−3·200 + 5·160]]
= (1/14) · [[480], [200]]
≈ [[34.3], [14.3]]
Although the inverse method provides clean, explicit solutions, it is not always the most efficient approach for computation. Students should develop good computational judgment.
| Method | Operation Count (Approximate) | Best Used When |
|---|---|---|
| Row reduce [A | b] | ≈ (2/3)n³ operations | Solving a single system; numerically safer for large n |
| Compute A⁻¹ then multiply | ≈ 2n³ + n² operations | Many systems with the same A; explicit formula needed |
In large computational problems, computing A⁻¹ explicitly can accumulate rounding errors. Modern numerical software (MATLAB, NumPy, Wolfram Alpha) solves linear systems using LU decomposition and backward substitution rather than explicitly forming the inverse. These methods are more numerically stable. Use inverses for conceptual understanding and for problems with repeated right-hand sides, but rely on row reduction or numerical methods for large-scale single systems.
Not true. Only invertible (nonsingular) matrices have inverses. Before computing, check whether det(A) ≠ 0, or verify that row reduction of [A | I] produces no zero row on the left.
The notation A⁻¹ means "the specific matrix such that AA⁻¹ = I." It is a label for the inverse, not a fraction. Matrices cannot be divided.
Correct: For AX = B, solution is X = A⁻¹B (left-multiply). For XA = B, solution is X = BA⁻¹ (right-multiply). These are different because matrix multiplication is not commutative.
The formula A⁻¹ = (1/(ad−bc))[[d, −b], [−c, a]] applies only to 2 × 2 matrices. For 3 × 3 or larger, use row reduction on [A | I].
Correct: (AB)⁻¹ = B⁻¹A⁻¹. The order is reversed — think of the socks-and-shoes analogy. To undo A then B, first undo B, then undo A.
After computing A⁻¹, always check that AA⁻¹ = I. This takes one matrix multiplication and catches arithmetic errors before they propagate into subsequent work.
If you only need to solve one system, computing the full inverse first is unnecessary extra work. Row-reduce [A | b] directly. Computing A⁻¹ costs roughly three times more operations.
If det(A) = 0 or row reduction produces a zero row, stop immediately. There is no inverse. Attempting to continue will produce meaningless results.
Throughout this chapter, the determinant has appeared as the key test: a matrix is invertible if and only if its determinant is nonzero. In Chapter 6, we will develop determinants fully. We will see that determinants measure how a linear transformation scales area (in 2D) and volume (in 3D), encode orientation — whether a transformation preserves or reverses spatial orientation — and provide a formula for computing inverses (Cramer's Rule).
Understanding determinants will complete our picture of matrix invertibility and prepare us for eigenvalues in Chapter 12, where determinants appear again in the characteristic polynomial.
The core question from this chapter — "Is a matrix invertible?" — will be answered in Chapter 6 by asking: "Is the determinant nonzero?" These two questions are equivalent, and Chapter 6 explains the geometric reasoning behind the determinant's remarkable power to encode area, volume, and structural collapse in a single number.
Find the inverse of each matrix, if it exists. State whether the matrix is invertible or singular.
Use row reduction to find the inverse:
Each solution below contains an error. Identify the error and write the correct solution.
What is the error in both the formula application and the result? What is the correct inverse?
Identify the error. What is the correct solution procedure and answer?
What is the error? State the correct rule and explain why order reverses.
What is the fundamental error here? What should be concluded and done instead?
Why is this impossible? Identify the structure that guarantees singularity before any row operations are needed.
Chapter 6 — Determinants will show exactly why det(A) ≠ 0 is the condition for invertibility. Determinants measure area and volume scaling, and the relationship to invertibility becomes clear through geometry. The formula for inverses using cofactors (Cramer's Rule) gives an explicit connection between determinants and the entries of A⁻¹.
Chapter 7 — Linear Transformations will show that invertible matrices correspond exactly to reversible transformations — bijections from ℝⁿ to ℝⁿ that preserve linear structure. Every concept from this chapter will reappear there in a richer geometric language.
Calculus and Multivariable Calculus use matrix inverses through the Jacobian matrix, which generalizes the derivative to vector-valued functions. The question "is the Jacobian invertible near a point?" determines whether coordinate changes and the Inverse Function Theorem can be applied.
Differential Equations use matrix exponentials and inverse matrices to build general solutions to systems of differential equations. Understanding A⁻¹ is prerequisite for these methods.
Data Science and Machine Learning use matrix inverses in regression (via the normal equations), in signal processing, and in the training and structure of linear models. The formula (AᵀA)⁻¹Aᵀb that arises in least squares depends directly on invertibility.