3
$\begingroup$

Let $n\in \mathbb N.$ Let $W$ be a non-trivial subspace of $n\times n$ symmetric matrices such that for every $x\in \mathbb R^n\setminus \{0\}$ there exists $a_x\in \mathbb R^n\setminus \{0\}$ such that $x\otimes a_x+a_x\otimes x\in W.$ Then prove that $\dim W\geq n.$

Notation: For $a,b\in \mathbb R^n$, $a\otimes b+b\otimes a=ab^T+ba^T$ is an $n\times n$ symmetric matrix.

We can prove this till $n\leq 3$ by contradiction.

For $n=1,$ it's simple because dimension of $1\times 1$ symmetric matrices is $1$.

For $n=2$, dimension of $2\times 2$ symmetric matrices is $3$. So, if $\dim W\leq n-1=1$ then for the canonical basis $\{e^1,e^2\}$ of $\mathbb R^2$ there exist $p^1,p^2\in \mathbb R^2\setminus \{0\}$ such that $e^1\otimes p^1+p^1\otimes e^1, e^2\otimes p^2+p^2\otimes e^2\in W$ and $\{e^1\otimes p^1+p^1\otimes e^1, e^2\otimes p^2+p^2\otimes e^2\}$ is liearly dependent. From there after some calculations, we got that $W=span \{e^1\otimes e^2+e^2\otimes e^1\}.$ After that I took the element $e^1+e^2$ and for this element there exists $p\in \mathbb R^2\setminus \{0\}$ such that $(e^1+e^2)\otimes p+p\otimes (e^1+e^2)\in W$. As $\dim W\leq 1,$ the set $\{(e^1+e^2)\otimes p+p\otimes (e^1+e^2), e^1\otimes e^2+e^2\otimes e^1\}$ is linearly dependent. From here also we got a contradiction that $p=0.$ Hence $\dim W\geq 2.$

For $n=3,$ similarly I considered three elements $e^1\otimes p^1+p^1\otimes e^1, e^2\otimes p^2+p^2\otimes e^2, e^3\otimes p^3+p^3\otimes e^3$ and taking the set $\{e^1\otimes p^1+p^1\otimes e^1, e^2\otimes p^2+p^2\otimes e^2, e^3\otimes p^3+p^3\otimes e^3\}$ as linearly dependent, I again got contradiction (It reduces into several cases, subcases and much more calculations than for $n=2$).

For $n\geq 4$, I was trying in a similar way but the calculations were getting much more difficult and I can not find any other way to prove for the general case.

Any help is appreciated. Thank you.

$\endgroup$
8
  • $\begingroup$ Your argument trivially gives a bound of $\dim(W) \geq n-1$. Not sure if that’s useful. It’s just the one extra dimension that is tricky. $\endgroup$ Commented Aug 4, 2023 at 15:59
  • 1
    $\begingroup$ @NathanielJohnston: Could you explain this in more detail, I don't see how to get $n-1$ easily. Just listing rank $2$ operators by going through a basis with $x$ might not be enough since the operators could come in pairs, with repetitions: for example $(x,a)=(e_1,e_2)$, then $(x,a)=(e_2,e_1)$ etc., so when we're done, we have $n/2$ linearly independent operators. $\endgroup$ Commented Aug 5, 2023 at 18:16
  • $\begingroup$ @ChristianRemling: you’re right, we just get n/2. I was just doing naive row/column counting, but forgot to take into account duplicates properly. $\endgroup$ Commented Aug 5, 2023 at 18:43
  • $\begingroup$ I think one can fight one's way through this by induction, but it probably won't be fun. I think it's best to try to prove the more general version: if for all $x\in M$, for some subspace $M$, ..., then $\dim W\ge \dim M$. If $T(x,a)$ denotes the rank $2$ operator from the OP and $P$ is a projection, then $P(T(x,a))P=T(Px,Pa)$, so one can restrict to subspaces as long as there are $a$'s available that are not orthogonal to this subspace. I would like to claim that I got close to a proof along these lines, but couldn't make the details work, and I'm giving up now. $\endgroup$ Commented Aug 5, 2023 at 19:10
  • $\begingroup$ @ChristianRemling: I was first taking an element $x\otimes a_x +a_x\otimes x$ and after that another $y\in \mathbb R^n\setminus \{0\}$ such that $y\in \{x,a_x\}^{\perp}$ and this way I was getting another element $y\otimes a_y+a_y\otimes y$ and I was able to show that these two elements are linearly independent. But constructing this way , it's becoming difficult to show the linear independence of the elements. But it's working upto $n=5$ to get $n-1$ linearly independent elements. $\endgroup$
    – mathew
    Commented Aug 8, 2023 at 9:38

0

You must log in to answer this question.

Browse other questions tagged .