7
$\begingroup$

There is the standard proof using $$\det(A)=\det( A^{T} ) = \det(-A)=(-1)^n \det(A)$$ I would like a proof that avoids this. Specifically, there is the proof that for $A$ a $\bf{real} $ matrix, the transpose is the same as the adjoint, which gives (using the complex inner product) $\lambda \|x\|^2 =\langle Ax, x \rangle= \langle x, -Ax \rangle=-\overline{\lambda } \|x\|^{2}$, so any eigenvalue is purely imaginary. Then we conclude that, since any odd-dimensional real matrix has a real eigenvalue, that eigenvalue must be zero. This argument doesn't work for a general complex skew-symmetric matrix. Is there something I'm missing, is there a way to modify this argument to get that zero is an eigenvalue for the complex case? Also, can somebody please give a geometric reason why odd-dimensional skew-symmetric matrices have zero determinant (equiv., a zero eigenvalue)?

Thanks!

$\endgroup$
  • $\begingroup$ What do you mean by skew-symmetric? Literally skew-symmetric, or antihermitian? $\endgroup$ – tomasz Jul 1 '12 at 0:57
  • $\begingroup$ Literally skew-symmetric, the transpose equals negative. $\endgroup$ – FGerard Jul 1 '12 at 5:11
8
$\begingroup$

Here is a "geometric" argument for the real case. The skew-symmetric condition is equivalent to $\langle Av, v\rangle =0$ for all vectors $v$. Geometrically, $Av$ is orthogonal to $v$. If $Av$ is never zero for $v\ne 0$, then we get a nonvanishing vector field on the unit sphere, contradicting the Hairy Ball theorem.

$\endgroup$
  • $\begingroup$ That's very nice - I've never seen that connection before! $\endgroup$ – Jason DeVito Jul 1 '12 at 1:01
  • $\begingroup$ This is great, pretty much along the lines of what I wanted for a geometric idea. Do you have an idea of how get the result for the complex case without using determinant (or using it in a different way)? $\endgroup$ – FGerard Jul 1 '12 at 5:12
  • $\begingroup$ @FGerard I don't. Do you have a geometric interpretation of the condition $A^T=-A$ in the complex case? $\endgroup$ – user31373 Jul 1 '12 at 19:53
  • $\begingroup$ @LeonidKovalev I do not. I guess this was actually two questions: one was if someone could show me a proof of this general fact, using 'adjoint $\endgroup$ – FGerard Jul 2 '12 at 16:36
  • $\begingroup$ - ' properties. The second question was if somebody could give a geometric or topological idea of what was going on. I read that skew-symmetric matrices for the Lie Algebra for O(n) (in the real case ?), so perhaps this property for odd-dimensions might be explained from that perspective. $\endgroup$ – FGerard Jul 2 '12 at 16:39
1
$\begingroup$

The proof you wrote down works in any field of characteristic not equal to $2$. Here is a similar such proof which avoids determinants due to Ian Frenkel.

Pass to the algebraic closure. Find $P$ such that $A = PDP^{-1}$ with $D$ upper triangular (for example using Jordan normal form, but you don't need to work this hard; it follows from the existence of eigenvectors). Then the diagonal entries of $D$ are the eigenvalues of $A$. Moreover, $$A^T = (P^T)^{-1} D^T P^T$$

where $D^T$ is lower triangular, so the diagonal entries of $D$ are also the eigenvalues of $A^T$. Since $A^T = -A$ it follows that the multiset of eigenvalues of $A$ is closed under negation. Since there are an odd number of them there must be an eigenvalue which is its own negative, so it must be zero.

$\endgroup$
  • $\begingroup$ I am not completely sure whether I need determinants to show that the multiset of eigenvalues is well-defined in the case that $A$ has nontrivial Jordan blocks, so run this proof for $A$ having distinct eigenvalues and then observe that such matrices are Zariski dense so the conclusion holds in general. $\endgroup$ – Qiaochu Yuan Jul 8 '12 at 6:11

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.