TITLE: The Hausdorff dimension of the union of singular orbits and exceptional orbits
QUESTION [5 upvotes]: Suppose we have a compact connected Lie group $G$ acting as isometries on a compact manifold $M^n.$ Then is it necessarily true that the Hausdorff dimension of the union of singular and exceptional orbits are no larger than the Hausdorff dimension of individual principal orbits? If not so, what about the union of singular orbits only?
It's known back in the 1950s through the works of Montgomery and Yang that in some cases, $\dim M^n\setminus M_{\text{principal}}\le n-2,$ the dimension of the complement of the union of principal orbits is at most $n-2.$ However, I'm not sure if they refer to Hausdorff dimension here. A search in the literature seems to indiate that there is no direct result on controlling $\dim M^n\setminus M_{\text{principal}}$ by the dimension of principal orbits.

REPLY [2 votes]: According to the paper "ORBITS OF HIGHEST DIMENSION" by Montgomery and Yang, by "singular orbits" one means orbits that have dimension less than the generic (or principle) ones.
For this reason, the answer to your question is rather negative. Indeed, take the round $S^n$ and consider on it the action of $S^1$ that fixes pointwise a geodesic $S^{n-2}$. Then the dimension of a generic orbit is $1$, but the union of singular orbits is the fixed $S^{n-2}$, i.e., it has dimension $n-2$.
PS. There is also a way to use this example to get a similar cohomogenity $4$ counter-example. Namely consider $S^4\times S^n$ and take the group $S^1\times SO(n+1)$ that acts diagonally. The first $S^1$-factor fixes $S^2\subset S^4$ (as previously) and the second acts transitively on $S^n$. Then a generic orbit has dimension $n+1$, but $S^2\times S^n$ is composed of obits of dimension $n$.