4

Let $G$ be a connected, closed subgroup of $\operatorname{GL}(n,\mathbb{C})$ and let $g \in G$. Is there a continuous function $f:[0,1] \to G$ such that $f(0) = g$ and $f(1)=1$ and $f(t) \cdot g = g \cdot f(t)$ for all $0 \leq t \leq 1$?

I believe if $\mathbb{C}^n$ has a basis of eigenvectors of $g$, then such an $f$ exists, but I am unsure in general.

Jack Schmidt
  • 55,589
  • Sorry, I confused it with the Riemannian exponential map. – Mikhail Katz May 14 '14 at 16:36
  • I think my group is "non-compact" which makes the exponential map no longer be surjective. – Jack Schmidt May 14 '14 at 16:37
  • Of course. Note that if you had such a family you could extend it to a 1-parameter group. I would assume that whenever the exponential map is not surjective, also there will be a counterexample, namely, the element that's not hit by the exponential map. – Mikhail Katz May 14 '14 at 16:38
  • Restatement: let $G$ be isomorphic to a closed connected subgroup of $GL(n,\mathbf{C})$. Is it true that every $g\in G$ belongs to the unit component of its centralizer $C_G(g)$? – YCor May 17 '14 at 11:58

1 Answers1

0

This answer tries to work out some hints, and they do work (thanks Yves!).

For very complicated reasons $f$ is quite likely to be (a composition of a homeomorphism of [0,1] and) the exponential map: $C_G(g)$ has the structure of a Riemannian manifold, and obviously the fastest way to get from $1$ to $g$ is a geodesic, which are given by the Riemannian exponential, which happens to also be the Lie exponential.

This answer gives an example of matrix that is not in the image of the exponential map (quite possibly the wrong exponential map, but I'll just hope everything works out) for the group $G=\operatorname{SL}_2(\mathbb{C})$:

$g = \begin{bmatrix} -1 & 1 \\ 0 & -1 \end{bmatrix}$.

The only matrices that commute with $g$ are those of the form $x(a,b) = \begin{bmatrix} a & b \\ 0 & a \end{bmatrix}$. This, $C_G(g) = \{ x(a,b) : a = \pm1, b \in \mathbb{C} \}$, is not a compact group, and is not path connected.

For $G=\operatorname{GL}_2(\mathbb{C})$, $f$ would be easy: $f(t) = x( 2t-1, 1-t )$. $f$ is continuous, $f(0)=x(-1,1) = g$, $f(1) = x(1,0) = 1$, and $f(t) \cdot g = g \cdot f(t)$ since $x(a,b) \cdot g = g \cdot x(a,b)$. However, for $G=\operatorname{SL}_2(\mathbb{C})$ we have $C_G(g)$ has two components, and $g$ does not lie in the identity component.

Jack Schmidt
  • 55,589
  • since you require $f$ to take values in $G$, your example is indeed a counterexample. 2) on the other hand, for every $g$ in $GL_n(\mathbf{C})$, its centralizer if $GL_n(\mathbf{C})$ is always connected, since it is the complement of a complex hypersurface in a complex vector space.
  • – YCor May 17 '14 at 12:04