Suppose we have an infinite compact (Hausdorff) group $G$, and a subgroup $H\leq G$ which is meagre.

Can $H$ always be covered by a countable family of nowhere dense sets $H_n$ such that $H_n^2$ is still nowhere dense, for each $n$?

Clearly, we can assume that $H$ is dense in $G$. I believe it holds when $H$ is (cointained in) a meagre $F_\sigma$ subgroup, because then we can just take for $H_n$ the closed nowhere dense sets which add up to $H$, and then use the fact that a meager closed set is nowhere dense.

On the other hand, just taking for $H_n$ a family of closed nowhere dense sets covering $H$ is not good enough -- for instance if any $H_n=H_n^{-1}$ is not null, its square will have nonempty interior by Steinhaus theorem.

I don't have much experience with abstract compact groups, so I have hard time even imaging $H,G$ which do not satisfy the assumptions of the previous paragraph...

  • $\begingroup$ Do you know the answer when G is the additive group of reals? $\endgroup$ – Ashutosh May 28 '15 at 14:51
  • $\begingroup$ @Ashutosh: The additive group of reals is not compact. If anything, a base example would be the circle group (or more generally, compact metrisable groups). And no, I don't know the answer even in this case. $\endgroup$ – tomasz May 28 '15 at 20:44
  • 1
    $\begingroup$ By reals i meant $2^{\omega}$. Here's a test situation to ponder: add $\aleph_1$ random reals and let H be the group generated by them. Can you write H as a countable union of sets whose self-sums are nowhere dense? $\endgroup$ – Ashutosh May 28 '15 at 22:17
  • 2
    $\begingroup$ I am glad to inform you that an international group of mathematicians is trying to answer your question. :-) We obtained negative answers under some additional axiomatic assumptions (for instance, Martin Axiom) and we are trying to obtain a negative answer in ZFC. Also we are trying to obtain a positive answer when the group $H$ is Borel. $\endgroup$ – Alex Ravsky Jun 2 '15 at 19:13
  • 1
    $\begingroup$ @AlexRavsky: Wow, thanks. Consistency of the negative answer is enough, as far as I am concerned. I might be interested in the special case where $H$ is analytic in the sense that it can be obtained from compact sets through the Souslin operation. Maybe this will prove a better assumption than Borelness? Borel sets tend to be rather odd in non-second-countable spaces. $\endgroup$ – tomasz Jun 2 '15 at 22:40

This problem was been answered in negative by M.Laczkovich (http://www.ams.org/journals/proc/1998-126-06/S0002-9939-98-04241-5/S0002-9939-98-04241-5.pdf). He constructed a proper Borel subgroup $H$ of the real line which cannot be covered by countably many sets $H_i$ with nowhere dense sums $H_i+H_i$.

On the other hand, Laczkovich proved that each non-open analytic subgroup $H$ of a Polish locally compact group $G$ can be covered by countably many closed sets of Haar measure zero.

Trying to generalize this result to non-locally compact groups, Laczkovich proved that any non-open analytic subgroup of a Polish group $G$ belongs to the sigma-ideal generated by the family $\mathcal F$ consisting of closed sets $A$ such that any non-empty open subspace of $A$ contains two relatively open non-empty sets $U,V$ with nowhere dense sum $U+V$ and difference $U-V$. Truly speaking, this result of Laczkovich is not quite satisfactory as each closed subset of $G$ containing a dense set of isolated points belongs to the family $\mathcal F$. Consequently, the $\sigma$-ideal generated by the family $\mathcal F$ coincides with the ideal of meager subsets of $G$.

But we can ask another problem: can each non-open analytic subgroup of a Polish Abelian group be covered by countably many closed Haar null subsets?

  • $\begingroup$ Thanks, this answers my question completely. :) (If not in the direction I had hoped for.) $\endgroup$ – tomasz Aug 26 '15 at 2:15
  • $\begingroup$ Actually, I withdraw my comment: the group of reals is not compact, just locally compact, so it does not really answer my question. Can this be easily fixed? $\endgroup$ – tomasz Aug 26 '15 at 12:21
  • $\begingroup$ To obtain a "compact" example, take the quotient of the real line by the cyclic subgroup $C$ generated by a point n the Borel subgroup $H$ constructed by Laczkovich. Then $H/C$ with a Borel subgrop of the compact group $R/C$ which does not admit the desired representation. $\endgroup$ – Taras Banakh Aug 26 '15 at 21:36
  • $\begingroup$ Right, of course. Well, you might as well take a quotient by ${\bf Z}$. I thought about that, but then took a glance at the paper and was mildly distressed with the use of convex hulls at the very beginning -- they would have been useless in the circle group. But of course this is locally just taking a countable union, so it preserves both borelness and meagreness of the subgroup, and if we had a "bad" (actually good) cover of the quotient, we could just pull it back to the reals, since meagreness is a local property. So thanks again. :) $\endgroup$ – tomasz Aug 27 '15 at 9:17

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.