13
$\begingroup$

Let $A$ be a square and singular matrix of order $n$.

Is $\operatorname{adj}(A)$ necessarily singular? What would be the rank of $\operatorname{adj}(A)$?

$\endgroup$
1
  • 2
    $\begingroup$ The title is a bit misleading, adjoint (most commonly meaning the complex conjugate of the tranpose) should be replaced by adjunct (the transpose of the comatrix). $\endgroup$
    – Max
    Mar 18, 2020 at 6:25

2 Answers 2

34
$\begingroup$

No, the adjugate of a singular matrix can be non-singular. But it happens only for the $1\times 1$ zero matrix.

Here is a complete classification, referring to this answer and this one. One always has $$\def\adj{\operatorname{adj}}A \cdot \adj(A) = \det(A) I_n.$$

  • If $A$ has rank$~n$, then it is invertible, and so is $\det(A)$, and $\adj(A)=\det(A)A^{-1}$ is invertible too, and has rank$~n$.

  • If $A$ has rank$~n-1$ then at least one $(n-1)\times(n-1)$ minor is nonzero, and so $\adj(A)\neq0$. On the other hand by the given relation the image of $\adj(A)$ is contained in the kernel of $A$ which has dimension$~1$ by rank-nullity; it follows that $\adj(A)$ has rank$~1$ in this case.

  • If $A$ has rank${}<n-1$ then all $(n-1)\times(n-1)$ minors are equal to zero, and so $\adj(A)$ has rank$~0$.

The cases where $\adj(A)$ has rank$~n$ are the first case for any$~n$, and the second case for $n=1$. (And, I would be inclined to say, the last case for $n=0$; but that of course cannot happen at all.) So the only case where $A$ is singular but $\adj(A)$ is not, is the case $A=(0)$ (with $n=1$).

$\endgroup$
5
  • 1
    $\begingroup$ If $A$ has rank $n-1$, how do we prove that there exists at least one cofactor matrix of $A$ with non zero determinant? That is, why is there some $i, j$ such that $A_{ij} \neq 0$? $\endgroup$
    – Anu
    Feb 16, 2018 at 12:27
  • 1
    $\begingroup$ Because one characterisation of the rank is that it is the largest size of a nonzero minor (with the $0\times0$ minor always being $1$, this is well defined). To see that such a nonzero minor indeed exists, let $r$ be the rank; select $r$ linearly independent columns (possible because $r$ is the column rank), which form a sub-matrix still of rank $r$, then select from it $r$ linearly independent rows (possible because $r$ is the row rank), they from a non-singular $r\times r$ sub-matrix. $\endgroup$ Feb 20, 2018 at 6:21
  • $\begingroup$ How the image of $adj(A)$ lies in the Kernal of $A$? Can you please explain? $\endgroup$
    – user464147
    May 28, 2019 at 5:09
  • 1
    $\begingroup$ @unknownx A matrix equation $P\cdot Q=0$ means that the image of$~Q$ is contained in the kernel of$~P$. This is an instance of that general fact, with $Q=\adj(A)$ and $P=A$. $\endgroup$ May 28, 2019 at 5:29
  • 1
    $\begingroup$ for the case where $A$ has rank $n-1$ you're assuming $n \ge 2$ ? $\endgroup$
    – BCLC
    Oct 23, 2021 at 18:11
5
$\begingroup$

Hint : $$A \cdot\mbox{adj}(A)=\det(A) I$$

If $\det(A)=0$, we get $A \cdot\mbox{adj}(A)=0$. Can $\mbox{adj}(A)$ be invertible?

For the rank, if you are familiar with linear transformations, prove that the above relation implies that the image of the transformation defined by $A$ must be in the kernel of the transformation defined by $\mbox{adj}(A)$. This yields an inequality of ranks.

$\endgroup$
5
  • $\begingroup$ If $|A| =0 $ then $ |A \cdot \text{adj}(A)| = |A| \cdot | \text{adj}(A)| =0$. Then I don't think it says anything about $|\text{adj}(A)|$? So how do we conclude if it is invertible or not? $\endgroup$
    – William
    Feb 7, 2022 at 21:01
  • 1
    $\begingroup$ @William you are making the problem too complicated by looking at determinants. You know that $A \cdot\mbox{adj}(A)=0$.. Now, if $AB=0$ and $B$ is invertible, what does this tell you about $A$? $\endgroup$
    – N. S.
    Feb 8, 2022 at 1:12
  • $\begingroup$ That $A=0$ is the zero matrix? I don't follow? $\endgroup$
    – William
    Feb 8, 2022 at 8:43
  • 1
    $\begingroup$ Oh if $A=0$ then $\text{adj}(A)=0$ hence non invertible. If $A ≠0$ then $\text{adj}(A)$ can't be invertible because that would imply $A=0$, hence a contradiction. Is that it? Have I done it? $\endgroup$
    – William
    Feb 8, 2022 at 8:45
  • $\begingroup$ @William Exactly. $\endgroup$
    – N. S.
    Feb 8, 2022 at 14:52

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .