1

Let $A=(a_{i,j})$ a square matrix whose elements are:

  • $0$ if $i=j$.
  • $1$ if $j>i$.
  • $-1$ if $j<i$.

Is there a simple way to find its rank?

Uri Toti
  • 533
  • 1
    Have you tried small examples? To what observation did they lead to? – Michael Hoppe Oct 15 '20 at 16:18
  • They seem to lead to the following conclusion:

    If the matrix is nxn, the rank will be n if n is even and n-1 if it's odd.

    – Uri Toti Oct 15 '20 at 16:19
  • first you work it out by hand for $n=1,2,3,4$ I think I will fiddle with this; there is what Newman, pages 56-60 calls the Skew Normal Form, related to the Smith normal form. The outcome is that, for a unimodular matrix $P,$ we construct $P^T A P = B,$ where the elements of $B$ are integers when $j = i+1$ and their negatives when $j=i-1,$ otherwise all zero. https://books.google.com/books?id=jEKFYutSGGYC&printsec=frontcover&source=gbs_ge_summary_r&cad=0#v=onepage&q&f=false – Will Jagy Oct 15 '20 at 16:20

3 Answers3

1

that was neat. Same rank as the matrix with entry $1$ when $j=i+1,$ then $-1$ when $j=i-1,$ otherwise $0$

Congruence:

$$ \tiny \left( \begin{array}{rrrrr} 1&0&0&0&0 \\ 0&1&0&0&0 \\ 0&-1&1&0&0 \\ 0&0&-1&1&0 \\ 0&0&0&-1&1 \\ \end{array} \right) \left( \begin{array}{rrrrr} 0&1&1&1&1\\ -1&0&1&1&1 \\ -1&-1&0&1&1 \\ -1&-1&-1&0&1 \\ -1&-1&-1&-1&0 \\ \end{array} \right) \left( \begin{array}{rrrrr} 1&0&0&0&0 \\ 0&1&-1&0&0 \\ 0&0&1&-1&0 \\ 0&0&0&1&-1 \\ 0&0&0&0&1 \\ \end{array} \right) = \left( \begin{array}{rrrrr} 0&1&0&0&0 \\ -1&0&1&0&0 \\ 0&-1&0&1&0 \\ 0&0&-1&0&1 \\ 0&0&0&-1&0 \\ \end{array} \right) $$

The final matrix is the Skew Normal Form in Newman, especially pages 56-60.

The process is now repeated with $E$ until the canonical form described by the theorem is obtained, and the procedure makes it clear that the rank of $A$ must be even.

Once the matrix is in this normal form, the new null (zero eigenvector)vector(s) becomes quickly apparent: none when $n$ even, exactly one when $n$ is odd, namely $(1,0,1,0,1)^T$ and multiplied by any constant.

Will Jagy
  • 139,541
1

checking the determinant, we see that $\det\big(A\big)= \det\big(A^T\big)= \det\big(-I_nA\big)= \det\big(-I_n\big)\cdot\det(A\big) = (-1)^n\cdot \det(A\big)$
so $n$ is odd$\implies$$A$ is singular.

Now it suffices to show that the matrix is invertible if $n$ is even. (When $n$ is odd, let $n':=n-1$ and consider the leading $n'\times n'$ principal submatrix -- if we show this is invertible, then $\text{rank}\big(A\big) = n'$.)

I suspect you're working over reals. Since all components of $A$ are integers, we may as well change the field to rationals, $\mathbb Q$ which doesn't change rank, ref Does real dimension equal rational dimension?

By assumption $n$ is even and it suffices to show $\det\big(A\big)\%2 \neq 0$. This is equivalent to changing the field to $\mathbb F_2$ and then calculating the determinant.

In $\mathbb F_2$ we have $A= - I+\mathbf {11}^T= I+\mathbf {11}^T$
(note: since $-1=1$ in $\mathbb F_2$, then $-I=I$)

By the matrix determinant lemma:
$$ \begin{align} &\det\big(A\big)\\ &= \det\big(I+\mathbf {11}^T\big)\\ &= \det\big(I\big)\cdot \big(1+\mathbf {1}^T(I)^{-1}\mathbf 1\big)\\ &= 1\cdot \big(1+\mathbf {1}^T\mathbf 1\big)\\ &=1\cdot \big(1+0\big)\\ &= 1 \end{align} $$
noting that $\mathbf {1}^T\mathbf 1 = \big(\sum_{k=1}^n 1\big) = 0 $ in $\mathbb F_2$ since $n$ is even.

Thus when $n$ is even, $\det\big(A\big)$ is odd, and therefore non-zero, which completes the proof.

user8675309
  • 10,034
  • they wanted the rank when the matrix is singular.. – Will Jagy Oct 15 '20 at 17:10
  • you are 12 seconds early -- I had bumped the submit button while still editing somehow – user8675309 Oct 15 '20 at 17:11
  • 1
    good work. I do the same thing, I get part of the answer typed up, see that it came out legible, then type the next section. Often I get downvotes in the middle – Will Jagy Oct 15 '20 at 17:13
  • it's brutal -- I think pressing 'return' will bring me to the next line to type more and yet somehow it becomes 'submit' – user8675309 Oct 15 '20 at 17:14
  • 1
    For the first few years I composed most answers in Latex on my home computer. I still do that for longer answers that have some interest for me. The tex file renders into about 300 pdf pages now; of course, I just make a pdf of the last few pages that have the current answer in progress. Easier for me to read and proofread than on MSE. – Will Jagy Oct 15 '20 at 17:17
  • I think there is a simple argument to show that a zero eigenvector has a very special form, forced on it by some kind of back substitution; anyway, when singular, the zero eigenspace has dimension one. – Will Jagy Oct 15 '20 at 17:35
  • Why $det(A^{T})=det(-I_nA)$? – Uri Toti Oct 15 '20 at 17:49
  • your $A$ has zeros on the diagonal, 1's above the diagonal and -1's below the diagonal. This means that $A$ is skew symmetric i.e. $A^T = -A= -I_n A$ – user8675309 Oct 15 '20 at 17:51
0

Here's a sneakier solution.
It suffices to prove $n-1\leq \text{rank}\big(A\big)$

$R:=A +\mathbf{11}^T$
$R$ has zeros below the diagonal so it is upper triangular and it has ones on the diagonal so $\text{rank}\big(R\big)=n$. Rank is sub-additive (ref: Show that $\operatorname{rank}(A+B) \leq \operatorname{rank}(A) + \operatorname{rank}(B)$) so we have

$n =\text{rank}\Big(R\Big) = \text{rank}\Big(A +\mathbf{11}^T\Big)\leq \text{rank}\Big(A \Big) + \text{rank}\Big(\mathbf{11}^T\Big)=\text{rank}\Big(A\Big) + 1$
$\implies n-1\leq \text{rank}\Big(A\Big)$

====
(i) When $n$ is odd, since $A=-A^T$, checking the determinant, we see that $\det\big(A\big) = 0$, so $n-1\leq \text{rank}\Big(A\Big)\leq n-1$, i.e. $\text{rank}\Big(A\Big) = n-1$.

(ii) Now for even $n$, the above tells us $\text{rank}\Big(A\Big)\in\big\{n-1,n\big\}$.
$\text{rank}\Big(A\Big)=n-1\implies A$ has an $n-1\times n-1$ principal submatrix that is invertible, see Prove the existence of a principal submatrix of order $r$ in $M\in\Bbb F^{n\times n}, M=-M^T,\ \operatorname{rank}(M)=r$ but this invertible $n-1\times n-1$ principal submatrix is odd dimensional and skew symmetric, so it must have a determinant of zero, a contradiction. Thus for even $n$, we know $\text{rank}\Big(A\Big)=n$

user8675309
  • 10,034