Idempotent Matrix: Definition, Properties & Examples

What Is An Idempotent Matrix?

An idempotent matrix is a square matrix that equals its own square, meaning $A^2 = A$. The word breaks into two Latin roots, idem (same) and potent (power), so an idempotent matrix keeps the same value no matter how many times you raise it to a power: $A^2 = A$, and therefore $A^3 = A$, and so on.

Two conditions sit inside that one line. The matrix has to be square, because a matrix can only be multiplied by itself when its columns match its rows. And the product $A \times A$ has to return the original grid, entry for entry.

The simplest members of the family are ones you already know. The identity matrix $I$ satisfies $I^2 = I$, and the zero matrix satisfies $O^2 = O$. Both pass the test without any work.

Throughout this article we use one running example of order 2 × 2:

$$A = \begin{bmatrix} 2 & -2 \ 1 & -1 \end{bmatrix}$$

It does not look special, yet squaring it gives back exactly what we started with. The next section shows that step by step.

How Do You Verify That A Matrix Is Idempotent?

To check whether a matrix is idempotent, multiply it by itself and compare the result to the original. If every entry matches, the matrix is idempotent; if a single entry differs, it is not.

Example 1: Test whether $A = \begin{bmatrix} 2 & -2 \ 1 & -1 \end{bmatrix}$ is idempotent.

Multiply $A$ by $A$ using the row-by-column rule, one entry at a time:

$$A^2 = \begin{bmatrix} 2 & -2 \ 1 & -1 \end{bmatrix}\begin{bmatrix} 2 & -2 \ 1 & -1 \end{bmatrix} = \begin{bmatrix} (2)(2)+(-2)(1) & (2)(-2)+(-2)(-1) \ (1)(2)+(-1)(1) & (1)(-2)+(-1)(-1) \end{bmatrix}$$

Work each entry:

$$A^2 = \begin{bmatrix} 4-2 & -4+2 \ 2-1 & -2+1 \end{bmatrix} = \begin{bmatrix} 2 & -2 \ 1 & -1 \end{bmatrix}$$

The result is identical to $A$, so $A^2 = A$.

Final answer: $A$ is idempotent.

There is also a shortcut for the 2 × 2 case. A matrix $\begin{bmatrix} a & b \ c & d \end{bmatrix}$ is idempotent exactly when $d = 1 - a$ and $bc = a - a^2$. Check the running example: $a = 2$, so $d = 1 - 2 = -1$, and $bc = (-2)(1) = -2$ matches $a - a^2 = 2 - 4 = -2$. Both conditions hold, which confirms the full multiplication above.

Which Matrices Are Idempotent?

Idempotent matrices come in more shapes than the identity. Any of the following passes the $A^2 = A$ test.

• The identity matrix. $I^2 = I$ for every size, so the identity is idempotent at every order.

• The zero matrix. $O^2 = O$, the trivial case that is easy to overlook.

• Projection matrices. A matrix that projects a vector onto a line or plane is idempotent, because projecting an already-projected vector leaves it where it is. The 2 × 2 matrix $\begin{bmatrix} 1 & 0 \ 0 & 0 \end{bmatrix}$ flattens every point onto the x-axis, and squaring it returns itself.

• Diagonal matrices with 0 or 1 on the diagonal. A diagonal matrix is idempotent only when each diagonal entry is 0 or 1, since $d^2 = d$ forces $d = 0$ or $d = 1$.

Larger idempotent matrices work the same way. This 3 × 3 example is idempotent, and its off-diagonal entries are far from zero:

$$M = \begin{bmatrix} 2 & -2 & -4 \ -1 & 3 & 4 \ 1 & -2 & -3 \end{bmatrix}, \qquad M^2 = M$$

Squaring $M$ reproduces all nine entries, which you can confirm with the same row-by-column method used above.

What Are The Properties Of An Idempotent Matrix?

Every idempotent matrix carries a fixed set of properties that follow directly from $A^2 = A$. They are worth memorising, because most exam questions test one of them rather than the definition itself.

Table: The defining properties of an idempotent matrix.

Three of these deserve a short proof, because seeing why they are true is the fastest way to remember them.

Eigenvalues are 0 or 1. Suppose $Av = \lambda v$ for a non-zero vector $v$. Multiply by $A$ again: $A^2 v = \lambda^2 v$. But $A^2 = A$, so $A^2 v = Av = \lambda v$. That forces $\lambda^2 = \lambda$, whose only solutions are $\lambda = 0$ and $\lambda = 1$.

The determinant is 0 or 1. Taking determinants of $A^2 = A$ gives $\det(A)^2 = \det(A)$, so $\det(A)$ is 0 or 1. When the determinant equals 1 the matrix is non-singular, and the only idempotent matrix that reaches it is the identity. Every other idempotent matrix has determinant 0.

$I - A$ is idempotent too. Expand the square: $(I - A)^2 = I - 2A + A^2 = I - 2A + A = I - A$. So the complement of an idempotent matrix is itself idempotent, a fact projections rely on. The trace then counts the eigenvalues equal to 1, which is exactly the rank, so the trace and the rank always agree.

How Is An Idempotent Matrix Different From Nilpotent And Involutory Matrices?

Idempotent matrices are one of three families defined by what happens when you raise the matrix to a power, and students routinely mix them up. A nilpotent matrix collapses to zero after enough multiplications, while an involutory matrix undoes itself in two.

Table: How idempotent, nilpotent, and involutory matrices compare.

A nilpotent matrix satisfies $A^k = O$ for some power $k$; the example above squares straight to the zero matrix. An involutory matrix satisfies $A^2 = I$, so it is its own inverse; the example is a reflection that returns to the start after two applications.

The three families meet in one place. The identity matrix is the single matrix that is both idempotent and involutory, since $I^2 = I$ and $I^2 = I$ describe the same thing. There is also a clean bridge: if $A$ is idempotent, then $2A - I$ is involutory, because $(2A - I)^2 = 4A^2 - 4A + I = I$.

Why Does A² = A Matter?

The equation $A^2 = A$ is not a curiosity invented for exams. It is the exact algebraic signature of a projection, and that single idea is why idempotent matrices appear across so many fields.

Think about what a projection does. A projection sends every point to its shadow on a line, a plane, or some larger subspace. The key feature is that once a point is already on that subspace, projecting it a second time does nothing.

• Projecting twice equals projecting once. If $P$ is the matrix that performs the projection, then applying it twice is the same as applying it once, which is precisely $P^2 = P$.

• The image is fixed. Points already living in the target subspace are left exactly where they are, so $P$ acts as the identity on its own output.

• The complement splits the space. Since $I - P$ is also idempotent, it projects onto the leftover directions, and every vector breaks cleanly into a projected part plus a remainder.

That is the whole reason $A^2 = A$ shows up so often. Any operation that keeps its own result unchanged when repeated is idempotent, and projection is the most common operation of that kind.

The Mathematicians Behind The Idempotent Matrix

The word "idempotent" is younger than the matrices it describes. It entered mathematics in the nineteenth century, when algebra was shifting from solving single equations toward describing whole systems of objects and the operations on them.

Two other figures shaped the surrounding story:

• George Boole (1815–1864, England) used the idempotent property in logic, where his equation $x^2 = x$ captured the idea that "the class of the class" is just the class, the algebraic seed of modern computing.

• Arthur Cayley (1821–1895, England) built the algebra of matrices themselves in 1858, defining how to multiply and invert them, which is the ground on which the idempotent property stands.

Where Are Idempotent Matrices Used In The Real World?

Idempotent matrices run quietly under any process that projects data onto a smaller space and then leaves it there.

• Statistics and regression. The hat matrix in least-squares regression is idempotent; it projects the observed data onto the line of best fit, and projecting the fitted values again returns the same fit.

• Computer graphics. Casting a 3D scene onto a 2D screen, or a shape onto its shadow, is a projection, so the transformation matrix satisfies $A^2 = A$.

• Machine learning. Dimensionality-reduction methods project high-dimensional data onto a handful of directions, and the projection step is idempotent by construction.

• Signal processing. Filters that isolate one component of a signal act as projections, keeping that component fixed once it has been extracted.

One equation, $A^2 = A$, connects a statistician's regression, an artist's shadow, and a learning algorithm's compression. The same structure keeps reappearing wherever a system needs to hold on to part of the information and discard the rest.

What Are The Most Common Idempotent Matrix Mistakes?

These four errors account for most lost marks on idempotent matrices, and each comes from collapsing the definition into something it only resembles.

Confusing An Idempotent Matrix With The Identity Matrix.

Where it slips in:

A student reads "multiplying by itself gives itself" and pictures only the identity, concluding that every idempotent matrix must be $I$.

Don't do this:

Do not assume idempotent means identity. The identity is the only invertible idempotent matrix, but singular ones like $\begin{bmatrix} 1 & 0 \ 0 & 0 \end{bmatrix}$ are idempotent as well.

The correct way:

Test the actual definition, $A^2 = A$. Most idempotent matrices have determinant 0 and are nowhere near the identity.

Confusing Idempotent With Involutory.

Where it slips in:

A student blurs $A^2 = A$ with $A^2 = I$, treating "squares to something simple" as one rule.

Don't do this:

Do not swap the right-hand side. $A^2 = A$ defines idempotent; $A^2 = I$ defines involutory, and they describe different matrices with different eigenvalues.

The correct way:

Read the right-hand side carefully. Idempotent returns the matrix itself, so its eigenvalues are 0 or 1; involutory returns the identity, so its eigenvalues are $+1$ or $-1$.

Assuming Every Diagonal Matrix Is Idempotent.

Where it slips in:

A student sees that diagonal matrices are easy to square and decides they must all be idempotent.

Don't do this:

Do not generalise from the identity. A diagonal matrix like $\begin{bmatrix} 2 & 0 \ 0 & 0 \end{bmatrix}$ squares to $\begin{bmatrix} 4 & 0 \ 0 & 0 \end{bmatrix}$, which is not the original.

The correct way:

Square each diagonal entry. A diagonal matrix is idempotent only when every diagonal entry is 0 or 1, since those are the only numbers that satisfy $d^2 = d$.

An Arithmetic Slip When Squaring The Matrix.

Where it slips in:

A student multiplies matching positions instead of rows against columns, so the "square" is wrong before the idempotency test even begins.

Don't do this:

Do not multiply entry by entry. Matrix multiplication is never element-wise, so $A^2$ is not the matrix of squared entries.

The correct way:

Use the row-by-column rule for every entry of $A^2$, then compare with $A$. One mis-multiplied term is enough to label an idempotent matrix as non-idempotent by accident.

Practice Problems On Idempotent Matrices

Work each one, then check the answer that follows.

1. Verify that $\begin{bmatrix} 1 & 0 \ 0 & 0 \end{bmatrix}$ is idempotent.
   (Answer: squaring it gives $\begin{bmatrix} 1 & 0 \ 0 & 0 \end{bmatrix}$, so $A^2 = A$. Idempotent.)

2. Is $\begin{bmatrix} 1 & 1 \ 0 & 1 \end{bmatrix}$ idempotent?
   (Answer: its square is $\begin{bmatrix} 1 & 2 \ 0 & 1 \end{bmatrix}$, which differs from the original. Not idempotent.)

3. Show that $\begin{bmatrix} 3 & -6 \ 1 & -2 \end{bmatrix}$ is idempotent.
   (Answer: $d = 1 - a$ gives $-2 = 1 - 3$, and $bc = -6 = a - a^2 = 3 - 9$. Both hold, and direct multiplication confirms $A^2 = A$.)

4. For $A = \begin{bmatrix} 2 & -2 \ 1 & -1 \end{bmatrix}$, find the trace and state the rank.
   (Answer: $\text{trace} = 2 + (-1) = 1$, so the rank is also 1.)

5. Confirm that $I - A$ is idempotent when $A = \begin{bmatrix} 1 & 0 \ 0 & 0 \end{bmatrix}$.
   (Answer: $I - A = \begin{bmatrix} 0 & 0 \ 0 & 1 \end{bmatrix}$, and its square is itself. Idempotent.)

6. What values can the determinant of any idempotent matrix take?
   (Answer: only 0 or 1, because $\det(A)^2 = \det(A)$.)

Where Should You Go Next After Idempotent Matrices?

Idempotent matrices open onto the parts of linear algebra that explain why they behave the way they do, and a few natural doors lead outward from here.

1. Eigenvalues. The 0-or-1 rule is an eigenvalue fact, so the eigenvalue toolkit is where the property comes alive.

2. Rank Of A Matrix. The trace-equals-rank result ties idempotent matrices to how much of the space a projection actually keeps.

3. Matrices And Determinants. The determinant test that separated the identity from every other idempotent matrix generalises here.

If your child is building these foundations, a live Bhanzu trainer teaches idempotent matrices through the projection idea they were built to model, in the Bhanzu algebra program.