2
$\begingroup$

$M$ is $n\times n$ real (or complex) matrix. Also $M$ is nilpotent of degree 2, i.e. $M^2=0.$

Question. How does $M$ look like?

I just calculated that $2\times 2$ matrix must have following form $$\begin{bmatrix} gh & \pm g^2 \\ \mp h^2 & -gh \end{bmatrix}.$$

I wanted to compute conditions on $3\times3,4\times 4$ and look for some pattern, but I thought that such problem should have been done long time ago. So

Or just reference request. Are there any sources that deals with this problem?

If one wants to know the origin, then this problem is related to this my unsolved problem.

PS. I added (homological-algebra) tag, cause of condition $M^2=0.$

1 Answers 1

3

Note that for any matrix $M$ satisfying $M^2 = 0$ and any invertible $S$, the matrix $N = SMS^{-1}$ satisfies $N^2 = 0$. Thus, it is common when answering a question like this to solve it "up to similarity". That is, provide a single element from each conjugacy class.

If we suppose that $M$ is in Jordan canonical form: every Jordan block is a block associated with zero, and the maximum size is $2$. That is, there is one conjugacy class for each distinct partition of the form $n = 2+2+\dots+2+1+1+\dots+1$.

If we instead suppose $M$ is in Weyr canonical form, then $M$ has the form $$ M = \pmatrix{ 0_{k \times k}&0&I_{k \times k}\\ 0&0_{(n-2k) \times (n - 2k)}&0\\ 0&0&0_{k\times k}} $$ With $k = 0,1,2,\dots,\lfloor n/2\rfloor$.


Similarly: for any polynomial $p(x)$, it is easy to find the $M$ satisfying $p(M) = 0$ up to similarity. One useful reference is chapter 3 of Horn and Johnson's Matrix Analysis.

  • 0
    This is very reasonable to make constrains up to some form, but in the linked problem I have two matrices $H,V$ such that $H^2=0$ and $V^2=0.$ Next step would be to see implications of condition $HV=VH.$ If I would apply your answer I will end up with $S_1HS_1^{-1}S_2VS_2^{-1}=S_2VS_2^{-1}S_1HS_1^{-1}.$ Which wouldn't help me much. I think I will have to compute $3\times 3, 4\times 4$ and look for the pattern anyway.2017-02-04
  • 0
    @FallenApart one nice trick that comes out of this is that we can assume without loss of generality that $H$ (or $V$) is specifically one of the matrices listed above.2017-02-04
  • 0
    Your observations are so bright, that I am stunned. I will sleep over with this ideas and I will try to make respond tomorrow.2017-02-04
  • 0
    @FallenApart that last suggestion was actually wrong, so I deleted it. In any case, I hope you question gets some more attention now.2017-02-04
  • 0
    I sincerely thank you for giving bounty to my question. But it was already downvoted (due to the title I guess). I will try to find more catchy one.2017-02-04
  • 0
    If you want you may have the honor to choose the proper one.2017-02-04
  • 0
    I did my best there. Honestly, I don't know why some people downvote. In any case, it might be useful to include (in the body of your question) to explain how you got from the question regarding the Bott-Chern Laplacian to this matrix question. That might attract the attention of the more pde-oriented folks on the site.2017-02-04
  • 0
    Thanks to your help I have a fresh idea how to tackle my original problem. You saved me planty of time, by leading to the fact that statement from bouted question is false. Thank you one more time.2017-02-05