Let $A$ a matrix $n\times n$ over $\mathbb R$. I'm trying to prove that $A$ is invertible $\iff\det A\neq 0$.

If $A$ is invertible, there is $B$ s.t. $AB=I$, and thus $$\det(A)\det(B)=\det(AB)=\det(I)=1,$$ and thus $\det A\neq 0$.

I have problem to prove the converse. Let $A$ s.t. $\det(A)\neq 0$. I would like to say that $$\det(A)\det(A)^{-1}=1.$$ I know that if $A^{-1}$ exist then $\det(A)^{-1}=\det(A^{-1})$, but since I have to prove that $A^{-1}$ exist, I can't use this formula... so how can I conclude ?

  • 3
    $\begingroup$ Do you know the formula for $A^{-1}$ in terms of $\mathrm{det}(A)$ and the cofactor matrix for $A$? $\endgroup$ Jun 24 '18 at 4:53
  • $\begingroup$ @AlexWertheim: No I don't. Is it possible to do without this tool ? $\endgroup$
    – user330587
    Jun 24 '18 at 5:04

To continue the proof using your approach:

Use the definition of $\det( )$ as the unique alternating multilinear function acting on columns of the matrix such that $\det(I) = 1$ (as opposed to the definition that gives an algebraic expression for $\det()$ in terms of permuations).

$\det(A) \ne 0 \implies $ columns of $A$ are linearly independent $ \implies A$ is full rank.

To get the first arrow, you can do proof by contradiction. So assume the columns of $A$ are dependent and $A_1 = \sum_{i=2}^n c_iA_i$ for some $c_i$. Then $\det(A_1, …, A_n) = \det(A_1 - \sum_{i=2}^n c_iA_i, A_2 + c_2A_2, …, A_n + c_nA_n) = \det(0, A_2 + c_2A_2, …, A_n + c_nA_n) = 0$.

This is the contradiction.

To get the second arrow, it's just the definition of rank.

So $A$ is surjective. By rank-nullity theorem, it is also injective. Therefore it is bijective, proving the existence of $A^{-1}$ as a function. Then it is a small step to prove that $A^{-1}$ linear and then you can complete the proof.


Well! You see $A^{-1}= Adj A / det(A)$ is also an important result which clearly proofs both cases.

  • $\begingroup$ Sorry but I neither don't know the formula nor what is Adj(A). Is there an other way ? $\endgroup$
    – user330587
    Jun 24 '18 at 5:05
  • $\begingroup$ Depending on what you are allowe to assume, we also habe that determinant of $A $ is non-zero iff determinant of row echelon form of $A $ (say $R $) is non-zero iff $A $ is invertible. Using elementary matrices $A=E_nE_{n-1}\dots E_1$, with each $E_i $ an elementary (and so invertible) matrix. $\endgroup$
    – AnyAD
    Jun 24 '18 at 5:07
  • 1
    $\begingroup$ This result is an expression of Cramer's rule. $\endgroup$
    – lhf
    Jun 24 '18 at 19:17

To elaborate on some of the other answers.

The determinant is an alternating multilinear form on the columns (or rows) of a matrix. That means that if you write your matrix as $(v_1, \dots, v_n)$ where $v_i \in \mathbf{R}^n$ are the columns of the matrix, then

  1. $\det$ is linear in each column

  2. If $v_i = v_j$ for any $i \ne j$ then $\det(v_1, \dots, v_n) = 0$.

These two properties imply that if $v_1,\dots,v_n$ are linearly dependent, then $\det(v_1,\dots, v_n) = 0$. For suppose that $v_1, \dots, v_n$ are linearly dependent. Then we can write one of the columns as a linear combination of the others, and for simplicity, we will suppose that the first column can be written as a linear combination of the others. Thus, let

$$v_1 = \sum_{i = 2}^n \alpha_i v_i. $$

Then, we have by 1. and 2.,

\begin{align*} \det(v_1,v_2,\dots,v_n) &= \det \left( \sum_{i = 2}^n \alpha_i v_i,v_2,\dots,v_n\right) \\ &= \sum_{i = 2}^n \alpha_i\det \left(v_i,v_2,\dots,v_n\right) \tag{by 1.} \\ &= \sum_{i = 2}^n 0 \tag{by 2.} \\ &= 0 \end{align*}

Thus if $A$ is not invertible, then the columns of $A$ are linearly dependent, so $\det A = 0$. This is the first proof.

For the second proof, in terms of elementary matrices, we know that there are 3 kinds of elementary row (or column) operations:

  1. Scale any row by a non-zero $\alpha \in \mathbf{R}$
  2. Swap any two rows
  3. Add $\alpha$ times one row to a different row

Each row operation can be written in terms of matrix multiplication. Moreover, if $E$ is the matrix that does the row operation, then $\det E$ is $\alpha$, in case 1; $-1$ in case 2; and $1$ in case 3. When we row reduce $A$, we multiply $A$ on the left by a sequence $E_1,E_{2},\dots, E_k$ of elementary matrices to get

$$ E_kE_{k-1}\cdots E_2E_1 A = \begin{pmatrix} I_r & 0 \\ 0 & 0 \end{pmatrix}$$

where $I_r$ is the identity matrix of size $r = \operatorname{rank}(A)$.

$$ \det(E_k)\cdots \det(E_1) \det(A) = \det\begin{pmatrix} I_r & 0 \\ 0 & 0 \end{pmatrix}.$$

We have $\det(E_k)\cdots \det(E_1) \ne 0$ since $\det(E_i) \ne 0$ for any $i$. If $\det(A) \ne 0$ then we have

$$ \det\begin{pmatrix} I_r & 0 \\ 0 & 0 \end{pmatrix} \ne 0$$

which is only possible if $r = n$. I.e. when $A$ has full rank.


Devendra Singh Rana's answer is the actual reason why the determinant is useful and why we have this characterization: the formula $(\mathrm{Com}A)^TA= \mathrm{det}(A)I_n$ shows that it suffices for $\mathrm{det}A$ to be invertible (hence for a field, $\neq 0$) for $A$ to be !

However in the specific case of $\mathbb{R}$ (or a field more generally) there is another proof that works great : let $f: M_n(\mathbb{R})\to \mathbb{R}$ be a multiplicative ($f(AB)= f(A)f(B)$) nonconstant function. Then $f(A)\neq 0$ if and only if $A$ is invertible.

Of course, applying this to $f=\mathrm{det}$ gives the desired conclusion.

How do we prove the claim ? Well in one direction, argue as you do that invertible matrices have nonzero $f$. For this you first have to show that $f(I_n)=1$. This follows from $I_nI_n = I_n$ and the fact that $f$ is nonconstant.

Thus you know that invertible matrices have nonzero $f$. Now this implies that "$f(A)\neq 0$" is invariant under equivalence of matrices (if $P,Q$ are invertible, then $f(PAQ) = 0 \iff f(A)=0$).

But two matrices are equivalent iff they have the same rank ! Now if $A$ is not invertible, its rank is $<n$ and so $A$ is equivalent to some nilpotent matrix $N$: but $f(N)^n= f(N^n) = f(0)$ and you can argue (because $0\times 0 = 0$ and $f$ is nonconstant) that $f(0)=0$, and so $f(N)^n=0$, $f(N)=0$, hence $f(A) = 0$.

So the fact that $\det$ characterizes invertibility can be seen either through the comatrix formula which I restated in the beginning, or through abstract properties of the function $\det$.


Depending on what you are allowed to assume, we also have that determinant of $A $ is non-zero iff determinant of row echelon form of $A $ (say $R $) is non-zero iff $A $ is invertible. Using elementary matrices $R=E_nE_{n-1}\dots E_1A$, with each $E_i $ an elementary (and so invertible) matrix.


Iff $\det(A)$ was $0$ then $0$ would be an eigenvalue of $A$, hence $A$ wouldn’t be invertible, see Show that a matrix $A$ is singular if and only if $0$ is an eigenvalue..

  • $\begingroup$ Why the downvotes? $\endgroup$ Jun 24 '18 at 18:57
  • $\begingroup$ Didn't downvote you, but I think you meant to say zero is an eigenvalue, not an eigenvector $\endgroup$
    – Mark
    Jun 25 '18 at 0:03
  • $\begingroup$ Of course, corrected. $\endgroup$ Jun 25 '18 at 6:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.