Let $G/H$ be an affine symmetric space with involution $\sigma$, and $\mathfrak{g}=\mathfrak{m}\oplus \mathfrak{h}$ the Cartan decomposition of its Lie algebra. We can identify $G/H$ and $\exp(\mathfrak{m})$. Given $p,q\in \exp(\mathfrak{m})$, I am wondering how to compute the holonomy of the curve formed by the geodesic segments, $e\rightarrow p$, $p\rightarrow q$, $q\rightarrow e$, $e$ being the identity. If I understand correctly, the holonomy should be given by a $Ad_k$ with $k\in H$. Can we compute $k$ in closed form from $p$ and $q$? I suspect it might be possible but I didn't manage.

Thank you for your help.


1 Answer 1


There is a general formula, essentially due to Cartan:

For each $p\in G/H$, let $\iota_p:G/H\to G/H$ be the geodesic inversion through $p$, i.e., the map that reverses all geodesics through $p$. Of course, ${\iota_p}^2$ is the identity. If $AB\subset G/H$ is a geodesic segment with midpoint $M$, then the map $R_A^B:G/H\to G/H$ defined by $R_A^B = \iota_M\circ\iota_A$ is a symmetry of $G/H$, i.e., it is left action by an element of $G$, that carries $A$ to $B$. Note that $R_A^B$ depends on the actual geodesic segment $AB$, not just on the endpoints $A$ and $B$.

Now, it turns out that the derivative of $R_A^B$ at $A$, i.e., $(R_A^B)'(A):T_A(G/H)\to T_B(G/H)$, is the result of parallel translating tangent vectors at $A$ along the geodesic segment $AB$ to tangent vectors at $B$. As a result, if $A$, $B$, and $C$ are points in $G/H$ such that $AB$ is a geodesic segment with midpoint $M$, $BC$ is a geodesic segment with midpoint $N$, and $CA$ is a geodesic segment with midpoint $P$, then the holonomy around the sides of the resulting geodesic triangle (in the order $A\to B\to C\to A$) is given by $$ R_C^A\circ R_B^C\circ R_A^B = \iota_P\circ\iota_C\circ\iota_N\circ\iota_B\circ\iota_M\circ\iota_A\,. $$ (Note that this map fixes $A$, so, as expected, it represents left action by an element of $H$.)

In case one has a reasonably explicit formula for the map $\exp:{\frak{m}}\to G$ and its inverse, one can work out an explicit formula for the holonomy in that case. For example, when $G = \mathrm{SO}(n{+}1)$ and $H=\mathrm{SO}(n)$, so that $G/H = S^n$, the unit vectors in $\mathbb{R}^{n+1}$, and $A,B\in S^n$ are not antipodal (i.e., $A+B\not=0$), one finds $$ R_A^B(Z) = Z +2\,(A{\cdot}Z)\,B - \frac{(A{+}B)\cdot Z}{1+A{\cdot}B}\,(A+B). $$ Using this and a little spherical trigonometry, one sees that, for a geodesic triangle with vertices $A$, $B$, and $C$ (no pair antipodal), then the holonomy of the geodesic path $A\to B\to C\to A$ is trivial unless $A$, $B$, and $C$ are linearly independent, spanning a $3$-plane $E\subset\mathbb{R}^{n+1}$, in which case, the holonomy is rotation about $A$ in the $3$-plane $E$ by an angle equal to the (signed) area of the spherical triangle with vertices $A$, $B$, and $C$ in the $2$-sphere $S^n\cap E$, as expected.


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.