A free action of $\mathbb{Z}/3\mathbb{Z}$ on a Riemannian manifold $(M, g)$ is called an equilateral action if for every $x\in M$ all three points of orbit of $x$ have the same distance from each other.

What is an example of a Riemannian manifold which does not admit such an action but it already admit a smooth free action by cyclic group of order $3$?

  • 2
    $\begingroup$ Are you assuming compactness or completeness? If you don't I think you can find uninteresting counter examples of the form $M=\Bbb{R}^2\setminus\{0,p, Ap, A^2p\}$ where $A$ is a matrix that is conjugate to a rotation by $2\pi/3$, $p$ is any nonzero point and the Riemannian structure is induced from that of $\Bbb{R}^2$. I would expect you can make this into a complete example by tampering with the metric. So maybe you should assume compactness? $\endgroup$ – Olivier Bégassat Mar 20 '20 at 23:42
  • 2
    $\begingroup$ $A$ is only conjugate to a rotation (i.e. satisfies $A^3=I_2$) but it's not an isometry. The Z/3Z action is that of $A$ on $M$. I'm removing two orbits from the action of $A$ on $\Bbb{R}^2$. $\endgroup$ – Olivier Bégassat Mar 21 '20 at 0:00
  • 1
    $\begingroup$ @OlivierBégassat for sure it works for the plane minus 4 points, none 3 of which form an equilateral triangle.If $f$ is a homeomorphism, it acts on the end compactification, which is a 2-sphere. If it fixes all end points, then this is a homeo of order 3 with exactly 5 fixed points and this doesn't exist. So it permutes 3 and fixes the last two. A limit argument shows that the point at infinity is fixed, and that the three moved ones form an equilateral triangle. $\endgroup$ – YCor Mar 21 '20 at 0:12
  • 3
    $\begingroup$ I guess you want a manifold with a free action of $\mathbb{Z}_3$ such that there is no equilateral action for ANY Riemanninan metric. (Otherwise the question has no sense.) $\endgroup$ – Anton Petrunin Mar 21 '20 at 5:25
  • 2
    $\begingroup$ @AntonPetrunin for any metric space an equilateral action makes sense (a free continuous action of $C_3$ in which every orbit is equilateral). So asking for a Riemannian manifold with no such action is perfectly meaningful. While your modified question is quite trivial: if there's a free $C_3$-action, you can average to get an invariant Riemannian metric. $\endgroup$ – YCor Mar 21 '20 at 7:17

Consider a torus with three handles, where one handle is much larger than the others, and with a smooth and free $\mathbb{Z}_3$ action which permutes the handles.

Let $\gamma$ be a small geodesic loop going through one of the small handles. Let $z\gamma$ be an image of $\gamma$ under the group action which goes through the large handle. Let $p$ be a point on $\gamma$ such that $zp$ is far out on the large handle.

Now the distance between $p$ and $zp$ is almost the diameter of the manifold, and there is no third point $z^2p$ which makes an equilateral triangle with them. So there is no smooth and equilateral $\mathbb{Z}_3$ action on the manifold.


Here's a compact example without boundary (a 2-torus).

Choose a topological disc $D$ on the two-torus $T$, and choose a Riemannian metric so that $D$ has very close Gromov-Hausdorff distance (say $\le 1$) to a segment of length $20$, while $T\smallsetminus D$ has diameter $\le 1$ (so metrically $D$ is predominant, while all the homotopic part lies in $T\smallsetminus D$. Also assume (just to fix ideas) that there exists $x_0\in D$ such that $d(x_0,D)=20$ (so $x_0$ is the "opposite tip" of the segment). Again to fix ideas, suppose that $x_0$ is part a geodesic segment $(x_t)_{0\le t\le 10}$ such that $\sup_{x\in T}\inf_{t\in [0,20]}d(x,x_t)\le 1$.

Long torus picture

(On the picture, the complement of $D$ is the little left tip, including the handle, and $x_0$ is at the right tip.)

Then $T$ (with this metric) has no large equilateral triangle: indeed if $r$ is the size of an equilateral triangle, such a triangle would be 1-close to a "triangle" $\{x_{t_1},x_{t_2},x_{t_3}\}$, with $t_1\le t_2\le t_3$ and $||t_i-t_j|-a\le 2$ for all $i\neq j$. So $t_3-t_1\le a+2$, $t_3-t_2,t_2-t_1\ge a-2$, hence $t_3-t_1\ge 2a-4$, so $2a-4\le a+2$, i.e., $a\le 6$.

Hence, if $f$ is an "equilateral" self-homeomorphism, we have $d(f(x),x)\le 6$ for all $x$. Let $B$ be the open $7$-ball around $x_0$. Then the open subset $U=B\cup f(B)\cup f^{-1}(B)$ is $f$-invariant, and contained in the $13$-ball around $x_0$. The connected component $U'\subset U$ of $x_0$ contains the $f$-orbit of $x_0$.

Let $T'$ be the quotient $T/\langle f\rangle$, so $T\to T'$ is a connected covering of degree 3. Since $\pi_1(U',x_0)\to \pi_1(T,x_0)$ is trivial, the covering is trivial in restriction to $U'$. We get a contradiction, since $U'$ is connected and contains as fiber the orbit of $x_0$.

  • 1
    $\begingroup$ I think you mean $a$ for the size of the equilateral triangle, and $x_0$ satisfies $d(x_0,T\setminus D)=20$. Very nice answer by the way, this is also how I read the OP's question. $\endgroup$ – Pierre PC Mar 22 '20 at 15:57

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.