Can A Non Square Matrix Be Invertible? | Inverses Made Clear

People call a matrix “invertible” when it can be undone. You apply the matrix, then apply its inverse, and you land back where you started. That idea fits when the matrix is square, like 3×3 or 5×5.

Once rows and columns don’t match, the map changes dimension. A rectangular matrix can still be useful, but it can’t be reversible in both directions the same way a square one can.

This article explains the strict meaning of inverse, why non-square matrices fail that test, and what to use instead when you want inverse-like behavior.

What “Invertible” Means In Linear Algebra

A matrix A is invertible if there is a matrix A⁻¹ such that A·A⁻¹ = I and A⁻¹·A = I. Both products must work. One product is not enough.

That double condition says the map is reversible both ways. Each input has one output, and each output comes from one input.

Britannica states this square-matrix definition directly in Britannica’s definition of an invertible matrix, which is a good anchor when the word “invertible” starts getting used loosely in notes or software.

Why A Non-Square Matrix Can’t Have A Two-Sided Inverse

Let A be an m×n matrix. If A had a two-sided inverse B, then A·B would equal I_m and B·A would equal I_n. Those identities are different sizes when m≠n.

Shapes force the contradiction: A·B is m×m, while B·A is n×n. A single B can’t satisfy both unless m equals n. So a non-square matrix can’t be invertible in the strict, two-sided sense.

A quick intuition helps. A wide matrix squeezes many input coordinates into fewer output coordinates, so distinct inputs can share one output. A tall matrix maps into a larger output space, so it can miss whole directions in that space. Either way, a full reverse map breaks.

Taking A Non Square Matrix Be Invertible? Question Seriously

If “invertible” means the classic A⁻¹ that works on both sides, the answer is no. The definition itself forces the matrix to be square.

If “invertible” means “can I undo it for my task,” then the right answer depends on shape and rank. That’s where left inverses, right inverses, and the pseudoinverse come in.

One-Sided Inverses: Left And Right

Rectangular matrices can still be reversible on one side. This is not a loophole. It is a different definition with a different promise.

Left Inverse For Tall Matrices

If A is m×n with m>n (tall), a left inverse is a matrix L such that L·A = I_n. Apply A, then L, and you get back the original n-dimensional input.

A left inverse exists only when A has full column rank, meaning its columns are linearly independent. No nonzero input can get squashed to the zero vector.

Right Inverse For Wide Matrices

If A is m×n with mm. This guarantees each output in m dimensions is reachable.

A right inverse exists only when A has full row rank, meaning its rows are linearly independent.

What You Do And Don’t Get With One-Sided Inverses

A left inverse lets you get back inputs from outputs that actually came from A. It does not promise you can hit each possible output in the larger space when A is tall.

A right inverse lets you hit each output, yet it won’t pick a single input when many inputs map to the same output. You need an extra rule to choose one.

Rank And Null Space: The Core Tests

Rank is the number of independent directions a matrix can carry through. For an m×n matrix, rank can’t exceed min(m, n). That cap explains why rectangular matrices can’t be reversible both ways.

For tall matrices, full column rank (rank = n) is the condition for a left inverse. For wide matrices, full row rank (rank = m) is the condition for a right inverse.

The null space shows why getting back can fail. If there is a nonzero x with A x = 0, then x and 0 map to the same output, so no rule can get back inputs in one way from outputs.

Inverse-Like Tools You Can Use In Practice

Even when strict inversion is off the table, you still need ways to solve A x = b or to “undo” a transformation in a sensible way. These tools come up most often.

Moore–Penrose Pseudoinverse

The pseudoinverse A⁺ works for any matrix shape. It gives a best-fit solution to A x ≈ b. In least-squares settings (more equations than unknowns), A⁺b returns the x that minimizes ‖A x − b‖. In underdetermined settings (more unknowns than equations), it can return the smallest-length solution among many.

Left Inverse Formula When Columns Are Independent

When A is tall with full column rank, AᵀA is square and invertible. A common left inverse is (AᵀA)⁻¹Aᵀ. This algebra explains why least squares feels close to “inverting” A while A is not square.

Right Inverse Formula When Rows Are Independent

When A is wide with full row rank, a common right inverse is Aᵀ(AAᵀ)⁻¹. It maps an output back to one chosen input. Other right inverses exist too.

Injective And Surjective: Another Clean Lens

You can translate the inverse question into two properties of a linear map. A map is injective when different inputs always give different outputs. A map is surjective when all outputs are reachable from at least one input.

For a square matrix, being injective and being surjective end up as the same condition: full rank. That is why the square case feels tidy.

For a tall matrix, injective is the realistic goal. You can keep inputs distinct when the columns are independent, and that lines up with having a left inverse. Surjective is not possible onto the whole larger output space, because the outputs live in an n-dimensional slice inside an m-dimensional space.

For a wide matrix, surjective is the realistic goal. You can reach all outputs when the rows are independent, and that lines up with having a right inverse. Injective is not possible, because more input directions exist than output directions, so some distinct inputs must collide.

Determinants Don’t Apply To Rectangular Matrices

Students often reach for determinants as a fast invertibility test. That test is valid only for square matrices, since determinants are defined for n×n matrices.

If a square matrix has determinant 0, it is singular, and no inverse exists. If the determinant is not 0, the matrix is invertible. When the matrix is non-square, there is no determinant to compute, so you must switch to rank and independence tests.

This is another reason the strict definition is tied to square matrices: the standard tools that certify invertibility, like determinants, are built for the square case.

Two Tiny Shape Checks You Can Do By Inspection

Say A is 3×2. It takes a 2-component input and produces a 3-component output. If the two columns of A are independent, then no nonzero input can land on 0, so inputs stay distinct. In that case, a left inverse L exists with L·A = I₂.

Now say A is 2×3. It takes a 3-component input and produces a 2-component output. Even if the rows are independent, many different inputs can land on the same output, since one degree of freedom has to disappear. In that case, a right inverse R can exist with A·R = I₂, yet getting back of the original input is not single.

These checks don’t require heavy computation. They flow from dimension and independence. Once you train your eye on shape and rank, most “inverse” questions become quick.

How Software Uses The Word “Inverse” With Rectangular Matrices

Some calculators and libraries will still offer an “inverse” button when you hand them a rectangular matrix. Under the hood, that is often the pseudoinverse, not a two-sided inverse.

That’s fine as long as you know what promise you are getting. A pseudoinverse is designed for solving and fitting, not for perfect undoing. It gives you a stable, consistent answer that matches rank limits and data noise.

Vocabulary Map: Terms That Show Up In Class And Tools

Use this table to translate phrasing across textbooks and software.

Term You’ll See	What It Means	What You Can Check
Invertible matrix	Square matrix with a two-sided inverse	Square shape and full rank
Nonsingular / singular	Square with inverse / square without inverse	Rank equals n / rank below n
Full column rank	Columns independent	Rank = number of columns; null space is {0}
Full row rank	Rows independent	Rank = number of rows
Left inverse	L with L·A = In	Tall (or square) with full column rank
Right inverse	R with A·R = Im	Wide (or square) with full row rank
Pseudoinverse	Generalized inverse for best-fit solving	Defined for any shape

Conditions Checklist For Rectangular “Reversal”

This table is the fast way to decide what’s possible from shape and rank alone.

Matrix Shape	What You Can Have	Condition That Must Hold
m×n with m>n (tall)	Left inverse L with L·A = In	Rank = n
m×n with m	Right inverse R with A·R = Im	Rank = m
m×n with m=n (square)	Two-sided inverse A−1	Rank = n
Any shape	Pseudoinverse A+ for best-fit solving	Always defined

Putting It To Work In Homework And Projects

If a problem asks for “the inverse” of a rectangular matrix, check the context. Many times the task is often asking for a left inverse, a right inverse, or a least-squares solution.

If the matrix is tall and its columns are independent, you can safely use left-inverse ideas. If the matrix is wide and its rows are independent, right-inverse ideas can make sense. If rank drops, shift to the pseudoinverse or to a best-fit method.

If you want a deeper course-style explanation of inverse matrices and identity products, MIT OpenCourseWare’s notes on inverse matrices are a solid reference.