This is a follow-up question to the one asked here:

Given a complex symmetric $2n\times2n$-matrix $A$, i.e., $A\in \mathbb{C}^{2n\times2n}$ with $A = A^T$. Is it possible, to block-diagonalize $A$ using a (complex) symplectic matrix $R$, i.e., is there always a matrix $R \in \mathbb{C}^{2n\times2n} $ with

$$ R^T J R = J \tag{1}$$ such that $$ R^T A R = \begin{pmatrix} 0 & D \\ D & 0 \end{pmatrix} \tag{1},$$ where $$ J = \begin{pmatrix} 0 & 1 \\ -1 & 0 \end{pmatrix}$$ and $D$ a diagonal matrix. If yes, what would be a good procedure to obtain $D$ and $R$.

some remarks:

  • so different from the former question, I assume a complex matrix $A$ and drop the requirement that the matrix is Hamiltonian
  • numerical tests on random matrices seem to suggest that this is always possible
  • I hope the solution to this problem is not too trivial. I did check some linear-algebra books and did not find the required result

1 Answer 1


This fails even for $n=1$. In this case, the matrix $$ A = \begin{pmatrix} 1&0\\0&0\end{pmatrix} $$ can't be diagonalized in the form that you want because it's not zero, yet its determinant vanishes.

There is a test for when this can be done, though. It's enough to have $JA$ be semi-simple (i.e., diagonalizable). In this case, the eigenvalues of $JA$ come in opposite pairs, and, if $\lambda_1,\ldots,\lambda_n,-\lambda_1,\ldots,-\lambda_n$ is a list of the eigenvalues of $JA$, then you can let $D$ be the diagonal matrix whose entries are $\lambda_1,\ldots,\lambda_n$.

The thing you want to think about is that the Lie algebra of the symplectic group is exactly the set of matrices of the form $JA$ where $A$ is symmetric, and you are trying to conjugate $JA$ into the Cartan subalgebra consisting of the diagonal matrices of this form via the adjoint representation of the symplectic group on its Lie algebra.

(It's a general fact that you can conjugate the semi-simple elements of the Lie algebra into a Cartan subalgebra via the adjoint action, but you can't conjugate the nilpotent elements into a Cartan subalgebra.)

  • 1
    $\begingroup$ Thank you for this simple counterexample (it is hard to sample a null set with random numbers). I still have one question: why is it that the eigenvalues of JA come in opposite pairs? (I guess this should be immediately obvious but sadly not for me) $\endgroup$
    – Fabian
    Commented Apr 23, 2014 at 18:29
  • $\begingroup$ @Fabian: One way to see it is to notice that $J^T = J^{-1}$, so $J(JA)^TJ^T$ has the same eigenvalues as $(JA)^T$, and hence the same eigenvalues as $JA$ itself. But if we expand things, we see that $J(JA)^TJ^T = JAJ^TJ^T = -JA$ (since $J^2 = -I$), so the spectrum of $JA$ and its negative coincide. $\endgroup$ Commented Apr 24, 2014 at 16:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.