Linear Algebra
4.1012 min read

The Invertible Matrix Theorem

For a square matrix A∈Rn×nA \in \mathbb{R}^{n \times n}, there are many equivalent conditions for invertibility — all 10 (or more) are mutually equivalent. Proving any one implies all the others.

Five "injective" conditions: the map is injective, nullity = 0, columns are linearly independent, Ax=0A\mathbf{x} = \mathbf{0} has only the trivial solution, AA has a left inverse.

Five "surjective" conditions: the map is surjective, rank = nn, columns span Rn\mathbb{R}^n, Ax=bA\mathbf{x} = \mathbf{b} always has a solution, AA has a right inverse.

For square matrices, all 10 are equivalent (by the Collapse Theorem). When these hold, the left and right inverses coincide — there is a unique two-sided inverse A−1A^{-1}.

Formal View

Theorem 4.10 (Invertible Matrix Theorem)
For A∈Rn×nA \in \mathbb{R}^{n \times n}, the following are all equivalent: \begin{enumerate} \item AA is injective \item Nullity(A)=0\text{Nullity}(A) = 0 \item Columns of AA are linearly independent \item Ax=0A\mathbf{x} = \mathbf{0} has only x=0\mathbf{x} = \mathbf{0} \item AA has a left inverse \item AA is surjective \item Rank(A)=n\text{Rank}(A) = n \item Columns of AA span Rn\mathbb{R}^n \item Ax=bA\mathbf{x} = \mathbf{b} has a unique solution for every b\mathbf{b} \item AA has a two-sided inverse A−1A^{-1} \end{enumerate}

Interactive Visualization

Invertible Matrix Theorem

Why This Matters

The Invertible Matrix Theorem is one of the most powerful unifying results in linear algebra — know one property, know all.

  • Testing invertibility: compute rank, check if =n= n — much cheaper than finding an inverse explicitly
  • Eigenvalue characterization: AA is invertible iff 0 is not an eigenvalue
  • In machine learning, invertible weight matrices enable backpropagation without information loss

Quiz

Question 1

A 4×44 \times 4 matrix has nullity 1. By the Invertible Matrix Theorem, it is:

Common Mistakes

  • Applying the Invertible Matrix Theorem to rectangular matrices — it only holds for square matrices.
  • Thinking you need to check all 10 conditions — checking any ONE suffices.