9
$\begingroup$

Let $\mathcal{M}_r$ be the set of $n \times m$ matrices over $\mathbb{R}$ or $\mathbb{C}$ of rank $r$. What is the Euler characteristic of $\mathcal{M}_r$? Can someone point me towards a reference for this calculation?

$\endgroup$
2
  • 6
    $\begingroup$ Over $\mathbb C$, this should be zero unless $r=0$ because the circle group acts freely on this manifold via multiplication by unit complex numbers. Maybe it's zero over $\mathbb R$ as long as $r\geq 2$ as well. $\endgroup$
    – Will Sawin
    Nov 7 '20 at 2:07
  • 2
    $\begingroup$ @WillSawin running the same argument over $\mathbb R$ shows that for $r> 0$, the Euler characteristic is even. $\endgroup$ Nov 7 '20 at 2:17
14
$\begingroup$

$\mathcal M_r$ can be described as a fiber bundle over the product $Gr(r,n) \times Gr(r,m)$ of the Grassmanian of $r$-dimensional subspaces in an $n$-dimensional vector space with the Grassmanian of $r$-dimensional subspaces in an $m$-dimensional vector space, where the fibers are all isomorphic to $GL_r$. (These are vector spaces over the field $\mathbb C$ or $\mathbb R$ as appropriate.) So its Euler characteristic is $$\chi( Gr(r,n) ) \chi( Gr(r,m)) \chi(GL_r).$$

Over the complex numbers $\chi ( GL_r(\mathbb C))=0$ for $r>0$, making the product zero, and for $r=0$ the space $\mathcal M_r$ is a point, with Euler characteristic $1$.

Over the real numbers, $\chi(GL_r(\mathbb R))$ may be calculated by observing that $GL_r (\mathbb R)$ maps to $\mathbb R^r - \{0\}$ by a fibration whose fibers are all $\mathbb R^{r-1}$-bundles over $GL_{r-1}(\mathbb R)$, so $$ \chi(GL_r(\mathbb R))=\chi(GL_{r-1}(\mathbb R)) \chi ( \mathbb R^r - \{0 \} ) $$ which is $0$ for $r \geq 2$ since $\mathbb R^2 - \{0\}$ has Euler characteristic zero.

So the product vanishes for $r \geq 2$ and is $1$ for $r=0$.

Finally, for $r=1$, $\chi(GL_1(\mathbb R))=2$, and $Gr(1,n) = \mathbb R \mathbb P^{n-1}$ which has Euler characteristic $\frac{ 1 + (-1)^{n-1} }{2}$, so $$\chi (\mathcal M_1) = \frac{ (1+ (-1)^{n-1} ) (1+ (-1)^{m-1} ) }{ 2} $$ and is always even, as Arun Debray noted.

$\endgroup$
1
  • $\begingroup$ Very nice answer. $\endgroup$ Nov 8 '20 at 2:30

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.