2
$\begingroup$

When this matrix is diagonalizable? ($a_i \in \mathbb{R}$) $$ \begin{pmatrix} &&&a_1\\ &&a_2&\\ &\ddots&&\\ a_n&&&\\ \end{pmatrix} $$ I think I should probably consider characteristic polynomial of this matrix, and if all roots are simple, then the matrix is diagonalizable.

UPD: Also I suppose that if $\forall a_i \neq 0$ then the matrix is diagonalizable, but I can't prove that.

The matrix is anti-diagonal, of course.

$\endgroup$
  • $\begingroup$ Is $\;a_{ij}=0\;$ for $\;i+j\neq n\;$ in your matrix? I mean, is this an antidiagonal matrix? $\endgroup$ – DonAntonio Apr 28 '16 at 19:29
  • $\begingroup$ It will help to treat the case $n=2$ first. $\endgroup$ – Henning Makholm Apr 28 '16 at 19:29
  • $\begingroup$ You can extract some information about the eigenvalues using the fact that the square of such a matrix is diagonal (so that the eigenvalues of the square are just the diagonal entries) and that $\det (A^2) = (\det A)^2$. $\endgroup$ – Travis Apr 28 '16 at 19:30
  • 1
    $\begingroup$ math.stackexchange.com/questions/1090364/… check this $\endgroup$ – thebooort Apr 28 '16 at 19:34
  • $\begingroup$ Conjugate by the appropriate permutation matrix. $\endgroup$ – Adam Hughes Apr 28 '16 at 19:36
1
$\begingroup$

Let $e_i$ be the standard unit vectors, $i=1\ldots n$. The two-dimensional subspaces spanned by $e_i$ and $e_{n+1-i}$, $i = 1 \ldots \lfloor n/2 \rfloor$ and (if $n$ is odd) the one-dimensional subspace spanned by $e_{(n+1)/2}$, are invariant under your matrix, so everything reduces to the two-dimensional case.

If $a_1, a_2 \ne 0$, $ \pmatrix{0 & a_1\cr a_2 & 0\cr}$ has two distinct eigenvalues $\pm \sqrt{a_1 a_2}$, therefore is diagonalizable. Of course if $a_1 = a_2 = 0$, you have the $0$ matrix which is diagonalizable. However, if one of $a_1$ is $0$ and the other is not, the matrix is not diagonalizable (the only eigenvalue is $0$, but the null space is one-dimensional).

Back to the general case: the matrix is diagonalizable unless for some $i$, one of $a_i$ and $a_{n+1-i}$ is $0$ and the other is not.

$\endgroup$
  • $\begingroup$ I can't understand the beginning of your solution, could you please clearify me? $\endgroup$ – AnatoliySultanov Apr 28 '16 at 19:54
1
$\begingroup$

Lemma 1 : $A$ is diagonalizable over $\mathbb C$ if and only if $A^2$ is diagonalizable and $\ker A =\ker A^2$

Here, $A^2$ is luckily a diagonal matrix, so $A$ diagonalizable over $\mathbb C$ if and only if $\ker A =\ker A^2$, that is to say, if and only if $\forall i, a_i=0\iff a_{n-i+1}=0$

Lemma 2: $A$ is diagonalizable over $\mathbb R$ if and only if $\ker A =\ker A^2$ and all the eigenvalues of $A^2$ are nonnegative.

Here, this translates as $\forall i, (a_i=0\iff a_{n-i+1}=0) \;\text{and } a_ia_{n-i+1}\geq 0$


These results are similar to those found here Conditions of diagonalizability of $n \times n$ anti-diagonal matrix

$\endgroup$
  • $\begingroup$ Could you explain me how to prove Lemma 2, please? $\endgroup$ – AnatoliySultanov Apr 28 '16 at 20:05
  • $\begingroup$ @AnatoliySultanov if the eigenvalues of $A^2$ are negative then the eigenvalues of $A$ are complex and it is clearly not real-diagonalizable. but for the converse you have to check also if the $P$ in $P D P^{-1}$ is a real matrix $\endgroup$ – reuns Apr 28 '16 at 20:09

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.