Let $G\subset \operatorname{GL}(n)$ be a matrix Lie group. I am curious about curves $\gamma(t) = g \exp(tv)$, where $g \in G$, $v \in \mathfrak{g}$, and $\exp(.)$ is the matrix exponential. If we equip $G$ with a bi-invariant metric, then geodesics take the form $\gamma(.)$ up to a constant. Of course, many matrix Lie groups cannot be equipped with a bi-invariant metric. I was wondering if there exists an affine connection under which all geodesics look like $\gamma(.)$. In other words, the manifold and matrix exponentials coincide.
-
1$\begingroup$ Thanks for asking this! I remember struggling with the same issue some 15 years ago. Back then someone explained it to me IRL but I forgot what the answer was. It is good to have it written somewhere on the internet in a findable place! $\endgroup$– VincentCommented Jan 9 at 9:55
-
$\begingroup$ @Vincent My pleasure! I am in the same boat as you were. $\endgroup$– Spencer KraislerCommented Jan 9 at 17:37
1 Answer
Yes, take the the trivial connection with respect to the left trivialisation of the tangent bundle. Then, all of your curves $\gamma$ are geodesics, but there are no further geodesics.
Some more details: the tangent bundle of every (matrix) lie group is trivial: consider a basis $e_1,..,e_n$ of you Lie algebra, and the globally well-defined frame $X_1,..,X_n$ determined by $X_k(g)=g e_k$, where the right hand side is just the matrix product (if you have a matrix Lie group). Then you can define a unique connection by declaring the frame to be parallel. Equivalently, this means that all Cristoffel symbols vanish identically. This gives you a flat connection (with non-trivial torsion unless the Lie algebra is commutative, ie the lie bracket vanishes). Then, for any $v$ in the Lie group, the vector field $g\mapsto gv$ is parallel. Hence, your curves $\gamma$ are all geodesic.
If you want a torsion free connection, just add half of the commutator as a global connection 1-form. As this is skew, you get the same geodesics, but by construction this connection has vanishing torsion.
-
2$\begingroup$ Can you elaborate on this process? I do not know what the trivial connection and left trivialization mean. Could you define a frame and the resulting Christoffel symbols? I will accept for further elaboration. $\endgroup$ Commented Jan 9 at 5:57
-
2$\begingroup$ @SpencerKraisler, just take any frame whose vector fields are left invariant, and set all the Christoffel symbols equal to zero. $\endgroup$ Commented Jan 9 at 15:02
-
$\begingroup$ Wow I did not know it was this simple. Is this the Cartan connection, or is that something else? $\endgroup$ Commented Jan 9 at 17:37
-
$\begingroup$ Just a simple remark to add: if a Lie group admits a bi-invariant pseudo-Riemannian metric then one-parameter groups are geodesics. In other words, the exponential map for the associated (bi-invariant) Levi-Civita connection coincides with the exponential map of the Lie group. $\endgroup$– user515519Commented Jan 24 at 21:55