9
$\begingroup$

This is a standard theorem in linear algebra over fields that is still true when working over an integral domain.

For a commutative unital integral domain $R$, let $A$ be an $n\times n$ matrix with entries in $R$. The system of linear equations $A\mathbf{x}=0$ has a non-zero solution if and only if $\det(A) = 0$.

The forward direction of this statement is not true though if we are working over a ring $R$ that is not an integral domain, a counter example being $$ R = \boldsymbol{Z}_6 \qquad A = \begin{pmatrix} 3 & 0 \\ 0 & 3 \end{pmatrix} \qquad \mathbf{x} = \begin{pmatrix} 2 \\ 2 \end{pmatrix} \,. $$

But what about the other direction? Does $\det(A) = 0$ imply that $A\mathbf{x} = 0$ has a non-zero solution over an arbitrary commutative unital ring? I've played around for a bit, and have yet to find a counterexample.

$\endgroup$
  • 2
    $\begingroup$ How do you define determinant over an arbitrary ring? $\endgroup$ – José Carlos Santos Aug 10 '18 at 15:22
  • $\begingroup$ @JoséCarlosSantos you're right. Let's limit this to commutative unital rings. Looking at the definition of $\det$, the unique alternating $R$-multilinear form that sends $I_n$ to $1_R$ ..., it does require commutative. But now I'm curious how badly things go if you try to define the (left-)determinate in terms of the usual algorithm, where you'd multiply each element of a row on the left by the $\det$ of the corresponding minor, if you don't have a commutative unital ring. $\endgroup$ – Mike Pierce Aug 10 '18 at 15:27
  • 3
    $\begingroup$ @MikePierce Determinants cannot be defined over noncommutative rings. There is a similar tool for division rings: en.wikipedia.org/wiki/Dieudonné_determinant $\endgroup$ – egreg Aug 10 '18 at 15:39
  • $\begingroup$ I found this where Sylvia's theorem are trying to prove core.ac.uk/download/pdf/82216288.pdf $\endgroup$ – Upasana Mittal Aug 10 '18 at 15:51
  • $\begingroup$ See this thread: math.stackexchange.com/questions/71740/… $\endgroup$ – Pierre-Yves Gaillard Aug 10 '18 at 23:36
7
$\begingroup$

Let $A$ be square of size $n \times n$, and $\det A=0$. Then $A\,\textrm{adj}(A)=0$ where $\textrm{adj}(A)$ is the adjugate of $A$. As long as $\textrm{adj}(A)\neq 0$ then taking some column of $\textrm{adj}(A)$ gives you $x$ with $Ax=0$.

But what if adj$(A)=0$? In this case all minors of $A$ vanish. Find a largest square submatrix of $A$ with nonzero determinant. We can assume that it's size $r$ by $r$ and fills the top left corner of $A$. Let $B$ be the top-left $r+1$ by $r+1$ submatrix of $A$. Then $\det B=0$ but adj$(B)\ne0$. Let $y$ be the $(r+1)$-th column of adj$(B)$ and $z$ be the column vector of height $n$ got by appending zeroes below $y$. Then $z\ne0$ (due to the non-vanishing of the top left $r$ by $r$ determinant) and I claim that $Az=0$.

The top $r+1$ entries of $Az$ are certainly zero. This is the identity $B\,$adj$(B)=0$. If we look at another entry, say the $s$-th then it is zero, essentially by replacing the bottom row of $B$ by the first $r+1$-th entries of the $s$-th row of $A$, and noting that the new matrix has zero determinant, as it is obtained from a submatrix of $A$ of size $r+1$ by $r+1$ by elementary row operations. As all submatrices of $A$ of this size have vanishing determinant, this completes the proof that $Az=0$.

$\endgroup$
  • 2
    $\begingroup$ Which $z$ do you get if you set $$A=\begin{pmatrix}1&0&0\\ 0&0&0\\ 0&0&0\end{pmatrix}\ ?$$ $\endgroup$ – Pierre-Yves Gaillard Aug 10 '18 at 19:31
  • $\begingroup$ You most certainly mean, "... taking some column of $\mathrm{adj}(A)$ gives you $x$ with $Ax=0$." I've edit your answer. $\endgroup$ – Mike Pierce Aug 10 '18 at 22:35
  • $\begingroup$ @Pierre-YvesGaillard Hope it's right now! $\endgroup$ – Lord Shark the Unknown Aug 11 '18 at 1:52
  • $\begingroup$ Very nice! I'd say that the $s$-th entry is zero because it is the determinant of the submatrix of $A$ of size $r+1$ by $r+1$ obtained by replacing the bottom row of $B$ by the first $r+1$-th entries of the $s$-th row of $A$. See also the references at the end of math.stackexchange.com/a/71755/660. $\endgroup$ – Pierre-Yves Gaillard Aug 11 '18 at 12:22
5
$\begingroup$

In a commutative ring (with 1), it turns out that, yes, $Ax=0$ has a non-trivial solution if and only if $\det(A)$ is either zero or a zero divisor. This follows from a theorem of McCoy.

Here is a link to a related question of my own:

MSE -- Do these matrix rings have non-zero elements that are neither units nor zero divisors?

MO -- https://mathoverflow.net/questions/77816/do-these-matrix-rings-have-non-zero-elements-that-are-neither-units-nor-zero-divi

A link to a paper with an account of McCoy's rank theorem: http://math.berkeley.edu/~lam/amspfaff.pdf

$\endgroup$
-2
$\begingroup$

Consider $\pmatrix{2&3\cr 2&3}$ in $\mathbb{Z}/6$ its determinant is $0$, and $2x+3y=0$ implies that $2x=3y$. But the multiples of $2$ are $\{0,2,4\}$ and the multiples of $3$ are $\{0,3\}$.

$\endgroup$
  • 4
    $\begingroup$ The solution can use the fact that $2$ and $3$ are zero-divisors though. So $\langle 3,2\rangle$ is a non-zero solution. This is just an example of the statement being true. $\endgroup$ – Mike Pierce Aug 10 '18 at 15:41

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.