1
$\begingroup$

If $S$ is real skew-symmetric matrix, then prove that $\det(I+S) \ge 1+\det(S)$,The equality holds if and only if $n\le 2$ or $n\ge3$ and $S=O$

Actually I only have trouble in proving that if and only if all the eigenvalues of $S$ is $0$ , then $S=O$.

$\endgroup$
3
  • $\begingroup$ Actually, your last claim is not true, see here. $\endgroup$ May 19, 2021 at 16:56
  • $\begingroup$ I can prove till that if $n\ge 3$, then all the eigenvalues of $S$ is 0(if and only if), i have no idea what to do next. $\endgroup$
    – Ross Ren
    May 19, 2021 at 17:05
  • $\begingroup$ @Dietrich Burde, i saw that "Specifically, every $2n\times 2n$ real skew-symmetric matrix can be written in the form $Q \Lambda Q^T$ where $Q$ is orthogonal" from wikipedia. $\endgroup$
    – Ross Ren
    May 20, 2021 at 8:05

2 Answers 2

3
$\begingroup$

I hope this helps :

Since skew-symmetric is normal, i.e commute with adjoint, you know that by complexification of the vector space, we have canonical form for those operator. For example for $f$ such that $f = -f^{*}$ (as for skew-symmetric) $\exists P^{-1}= P^t$ such that $$A = P^{-1}S P = \begin{pmatrix}\textbf{0} & 0 \\ 0 & \Box\end{pmatrix}$$

Where the first $0$ are $k$ zero relative to the eigenvectors of $0$ eigenvalue of $f$, and the other part are $j$ (indeces are not very important here) block $\Box_\mu = \begin{pmatrix}0 & \mu \\ -\mu & 0 \end{pmatrix}$ relative to the purely imaginary eigenvalues of $f$ (Here $\mu \in \mathbb{R}$ and this is very important).

That being sad, since $\det(P)^2 = 1$ (being $P$ orthogonal) playing with Binet :

$$\text{det}(Id+S) = 1\cdot \text{det}(Id+S) = \text{det}(P)^2 \text{det}(Id+S) = \text{det}(P^{-1})\text{det}(Id+S)\text{det}(P) = \text{det}(P^{-1}(Id+S)P) = det(P^{-1}Id P + P^{-1}SP) = \text{det}(Id+ A)$$

It should clear that $I_d+A = \begin{pmatrix}Id_k & 0 \\ 0 & \Box\end{pmatrix}$ where the blocks now turned into $B_\mu = \begin{pmatrix}1 & \mu \\ -\mu &1 \end{pmatrix}$

And since the matrix is block diagonal we have that $\text{det}(Id+A) = 1^k \prod\limits_{i=1}^j B_{\mu_i}$.

Note that having $\text{det}( B_{\mu_i}) = 1 + \mu_i^2$, we end up with $\text{det}(I_d+A) = \prod\limits_{i=1}^j (1+\mu_i^2)$.

It should be clear now that

$$\prod\limits_{i=1}^j (1+\mu_i^2) \geq 1+ \prod\limits_{i=1}^j \mu_i^2 = 1 + \prod\limits_{i=1}^j \text{det}(\Box_{\mu_i}) = 1 + 1^k\prod\limits_{i=1}^j \text{det}(\Box_{\mu_i}) = 1 + \text{det}(P^{-1}SP) = 1 + \text{det}(P)^2\text{det}(S) = 1+\text{det}(S)$$

Edit : I think if this is right you can easily find the special cases for $n$ by yourself.

$\textbf{Addendum :}$ Rewatching this "proof" it seems to me that the equality holds if and only if holds $\prod\limits_{i=1}^j (1+\mu_i^2) = 1+ \prod\limits_{i=1}^j \mu_i^2$. If you are familiar with the first product, you should know or you could prove that $\prod\limits_{i=1}^j (1+\mu_i^2) = \sum\limits_{J \subset[j]}\prod\limits_{i \in J} \mu_i^2$. From this equality you see that holds if and only if $\mu_i^2 = 0$ for every $i$ since we're dealing with real numbers, i.e $\mu_i = 0$ for all $i$. But if we remember who $\mu_i$ was, we get $S = 0$ since $0$ is the only matrix similar to the $0$ matrix.

$\endgroup$
6
  • $\begingroup$ Thanks a lot for your help. Actually your last formulas are enough to cover the proof and I've already solved that.The problem is, if the equality holds and $n\ge 2$, then all the $\mu_i=0$,which is all I have gone so far. Could you help me with the last step that $S=O$? Your assistance will be highly appreciated. $\endgroup$
    – Ross Ren
    May 20, 2021 at 10:08
  • $\begingroup$ @RossRen I'm sorry, could you rephrase the question ? I don't understand what you are trying to prove $\endgroup$ May 20, 2021 at 10:10
  • $\begingroup$ OK. Using $AX=\lambda X$ and your last formulas, I proved the main proposition.I only have trouble in the special case when the equality is achieved. I mean, how to prove $S=O$ when I only knew all the eigenvalues are zeros. $\endgroup$
    – Ross Ren
    May 20, 2021 at 10:16
  • $\begingroup$ @RossRen Look at the Addendum and tell me what you think $\endgroup$ May 20, 2021 at 10:37
  • $\begingroup$ I've got your points.Thanks a lot. And sorry again for my ambitious problem. I only knew that using congruent transformation, we can get $diag${$S,...,S,0,...,0$} where $S$ denotes\begin{matrix} 0 & 1 \\ -1 & 0 \end{matrix} , can you only use this to solve th e problem so that I could finish the proof before learning about orthogonal matrices. $\endgroup$
    – Ross Ren
    May 20, 2021 at 10:54
2
$\begingroup$

Actually I only have trouble in proving that if and only if all the eigenvalues of 𝑆 is 0 , then 𝑆=𝑂.

since OP is stuck on this point and does not want to use orthogonality, here's a different finish.

It is immediate that $S=\mathbf 0\implies $ all eigenvalues of $S$ are 0.

We need to prove the other direction, i.e. having all eigenvalues (considered in $\mathbb C$) of real skew symmetric $S$ be 0 $\implies S=\mathbf 0$

$\text{trace}\big(S^2\big) = \sum_{k=1}^n \lambda_k^2 = \sum_{k=1}^n 0^2=0$
but $S^2=-S^TS\implies 0= \text{trace}\big(S^TS\big)=\Big\Vert S\Big \Vert_F^2 \implies S=\mathbf 0$

$\endgroup$
1
  • $\begingroup$ Thanks a lot. Your answer completely solves my problem. $\endgroup$
    – Ross Ren
    May 21, 2021 at 1:21

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.