Let $M_n \subseteq SO(2n)$ be the set of real $2n \times 2n$ matrices $J$ satisfying $J + J^{T} = 0$ and $J J^T = I$. Equivalently, these are the linear transformations such that, for all $x \in \mathbb{R}^{2n}$, we have $\langle Jx, Jx \rangle = \langle x, x \rangle$ and $\langle Jx, x \rangle = 0$. They can also be viewed as the linear complex structures on $\mathbb{R}^{2n}$ which preserve the inner product.

I'd like to understand $M_n$ better as a topological space, namely an $(n^2-n)$-manifold.

$M_1$ is just a discrete space consisting of two matrices: the anticlockwise and clockwise rotations by $\pi/2$.

For $n \geq 2$, we can see that $M_n$ is an $M_{n-1}$-bundle over $S^{2n-2}$. Specifically, given an arbitrary unit vector $x$, the image $y := Jx$ must lie in the intersection $S^{2n-2}$ of the orthogonal complement of $x$ with the unit sphere $S^{2n-1}$. Then the orthogonal complement of the space spanned by $x$ and $y$ is isomorphic to $\mathbb{R}^{2n-2}$, and the restriction of $J$ to this space can be any element of $M_{n-1}$.

Since the even-dimensional spheres are all simply-connected, it follows (by induction) that $M_n$ has two connected components for all $n \in \mathbb{N}$, each of which is simply-connected. For instance, $M_2$ is the union of two disjoint 2-spheres: the left- and right-isoclinic rotations by $\pi/2$. The two connected components of $M_n$ are two conjugacy classes in $SO(2n)$; they are interchanged by conjugating with an arbitrary reflection in $O(2n)$.

Is [each connected component of] $M_n$ homeomorphic to a known well-studied space? They're each an:

$S^2$-bundle over an $S^4$-bundle over $\dots$ an $S^{2n-4}$-bundle over $S^{2n-2}$

but that's not really very much information; can we say anything more specific about their topology?

  • 5
    $\begingroup$ Your $M_n$ is the Riemannian symmetric space $\mathrm{SO}(2n)/\mathrm{U}(n)$. I believe that its topology is quite well studied from that point of view. $\endgroup$ Sep 18, 2020 at 17:43
  • $\begingroup$ Thanks! Yes, it looks like the space is called DIII in Cartan's classification of compact Riemannian symmetric spaces. (I'll accept this if you post it as an answer.) $\endgroup$ Sep 18, 2020 at 18:35
  • 1
    $\begingroup$ yes, this is class DIII, the two connected components are distinguished by the Pfaffian, which equals $\pm 1$; in the physics context this matrix $M_n$ is the scattering matrix of a superconductor with preserved time-reversal symmetry but broken spin-rotation symmetry. The sign of the Pfaffian then distinguishes topologically trivial from topologically nontrivial superconductors. $\endgroup$ Sep 18, 2020 at 19:30
  • 1
    $\begingroup$ In Morse Theory, Milnor has proved that $\mathrm{O}(2n)/\mathrm{U}(n)$ can be identified with orthogonal linear transformations $J$ with $J^2=-I$ i.e. complex structure. $\endgroup$
    – C.F.G
    Sep 18, 2020 at 19:34

1 Answer 1


Your $M_n$ is (two copies of) the Riemannian symmetric space $\mathrm{SO}(2n)/\mathrm{U}(n)$ (which is $DI\!I\!I$ in Cartan's nomenclature). Its topology is well-studied from that point of view.


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.