0
$\begingroup$

Edit: According to comment by Leo Monsaingeon I revise my question:

Is there a Riemannian metric on $M_n(\mathbb{R})$ for which the function $trace$ is a bounded function on every complete(whole) geodesic? that is for every complete geodesic $\gamma$, the restriction $trace$ to $\gamma$ is a bounded function? What about the same question if we replace $M_n(\mathbb{R})$ with $GL_n(\mathbb{R})$?

Remark By a complete geodesic we DO NOT mean exclusively a geodesic whose parameter $t$ can be varied in whole $\mathbb{R}$, but we mean $\{\gamma(t)\mid t\in I\}$ where $I$ is the maximal interval of definition of $\gamma$. So we mean the whole trajectory of a non parametrized geodesic.

$\endgroup$
6
  • $\begingroup$ Do you mean complete geodesics? Or just two-points geodesics between arbitrary matrices $M,N$? In the latter case the Hilbert structure induced by the Frobenius scalar product should do, right? $\endgroup$ Commented Jun 6, 2020 at 14:50
  • 1
    $\begingroup$ @leomonsaingeon I think it is the Eucliean metric so the geodesic $\begin{pmatrix}t&0\\0&0\end{pmatrix}$ has an unbounded trace, right? $\endgroup$ Commented Jun 6, 2020 at 14:53
  • 1
    $\begingroup$ well, not if you mean "geodesics between arbitrary pairs of points", in which case your time parameter cannot run across the whole real line. But I guess from your comment that you really mean "complete geodesics", right? If so perhaps it would be worth editing your question (also, there is a typo in your title "o"->"of") $\endgroup$ Commented Jun 6, 2020 at 14:54
  • $\begingroup$ @leomonsaingeon I mean complete geodesic, otherwise not only trace but also every continuous function on a compact set9geoesic from M to N) is a bounded function $\endgroup$ Commented Jun 6, 2020 at 14:55
  • $\begingroup$ right, but the continuity of the trace function depends on the choice of the metric (although I agree that pretty much all the Riemannian structures should give an equivalent topology). So it was not completely clear for me what you meant. Now it is ;-) $\endgroup$ Commented Jun 6, 2020 at 14:57

1 Answer 1

3
+50
$\begingroup$

It is possible to find such metrics. However, they are a bit unnatural, in that they aren't really constructed from any inherent properties of matrices. For concreteness, I'm going to restrict my attention to $M_2$ and $Gl_2$, but this construction can be modified to work for any larger $n$ as well.

We consider $2 \times 2$ matrices $$ \left[ \begin{array}{cc} a & b \\ c & d \\ \end{array} \right]. $$

To construct the desired metric, we start by changing coordinates on $M_2$ as follows.

$x = a+d$, $y= \exp(a-d)$, $z=b$, $w=c$

This gives a map from $M_2$ to a half-space in $\mathbb{R}^4$. For a Riemannian metric, we use $$ds^2 = \frac{dx^2+dy^2}{y^2} +dz^2+dw^2.$$

Notice that the $x$-coordinate is simply the trace of the original matrix, so what's left to show is that $x$ is bounded on any geodesic. Different geodesics will have different bounds, however.

To show that $x$ is bounded on any geodesic, observe that as a Riemannian manifold, $(M_n, g)$ is simply the metric product $\mathbb{H} \times \mathbb{R}^2$. As such, the geodesics are of the form $\gamma(t) = (c(t), \ell(t))$. Here, $c(t)$ forms an arc of a semi-circle in $\mathbb{H} $ whose center is on the axis $y=0$. Meanwhile, $\ell(t))$ is a line in $\mathbb{R}^2$. As such, it follows the $x$ coordinate is bounded by the center and radius of $c(t)$.

When we try to do this for $Gl_2$, we have to deal with the fact that the space is disconnected. However, we can use a similar idea on each connected component. Namely, we change coordinates to $x = a+d$, $y= \exp(a-d)$, $z=b$, $w=ad-bc$ and take the metric $ds^2 = \frac{dx^2+dy^2}{y^2} +\frac{dz^2+dw^2}{w^2}.$ This gives $Gl_2$ the metric structure of $\mathbb{H} \times \mathbb{H}$ on each of its connected components, which also has the desired geodesic property.

$\endgroup$
0

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .