4
$\begingroup$

Let $A$ be a nonempty open connected subset of a (real) topological vector space $X$ such that $$2A-A \subseteq 8A$$ (for instance one could take $A=(-1,2)$).

Question. Is it true that there exists a nonempty open connected set $B\subseteq A$ such that $B$, in addition, is symmetric (i.e., $B=-B$)?

$\endgroup$
  • $\begingroup$ Have you proved it for $X=\Bbb R$ at least? $\endgroup$ – Adam Chalumeau May 15 at 10:16
  • 1
    $\begingroup$ @AdamChalumeau The hypotheses imply $0\in A$. $\endgroup$ – logarithm May 15 at 10:30
  • $\begingroup$ @logarithm Do they? Do you have a proof of this? If so, then $A \cap -A$ is a good choice for $B$. $\endgroup$ – Theo Bendit May 15 at 10:33
  • 1
    $\begingroup$ @PaoloLeonetti Those are the only connected open sets of $\mathbb{R}$, and of a general $A$ intersected with a line in a general tvs. $\endgroup$ – logarithm May 15 at 10:49
  • 1
    $\begingroup$ @Mirko The example was really supposed to demonstrate why $A \cap (-A)$ may not be connected, even though $0 \in A$ and $A$ is connected. An open example is easy enough too; just add (in the sense of Minkowski sum) an open ball of radius $1/3$ to $A$, and the intersection should still be disconnected (I don't want to compute the exact intersection). However, I agree that $2A - A \subseteq 8A$ (probably) does not hold here. Note that, if $0 \in A$, $A$ is open in a normed linear space, then $A$ always contains an open ball around $0$, which is open, connected, and symmetric. $\endgroup$ – Theo Bendit May 16 at 1:01
3
+50
$\begingroup$

I got many ideas from the comments, but I had to verify the details, and to modify and add some elements to convince myself that the answer is yes (as shown below).

So we have $\frac A8\subseteq \frac A4 - \frac A8\subseteq A$ (hence $\frac A8\subseteq A$, also $\frac A{64}\subseteq \frac A8$, and, by induction, $\frac A{8^n}\subseteq A$ for all $n\ge1$). Since $A$ is open, it is not difficult to see that if $C$ is the closure of $A$ then also $\frac A4 - \frac C8=\frac A4 - \frac A8\subseteq A$ (the details are provided in a corollary near the end). If $a\in A$ then the sequence $\frac a{8^n}$ converges to $0$ (I believe even in general TVS (yes, reference provided in a comment below by OP)), so $0\in C$. Hence $\frac A4=\frac A4-0 \subseteq\frac A4 - \frac C8=\frac A4 - \frac A8\subseteq A$. So we have:

(i) $\frac A{32} \subseteq \frac A4$ because $\frac A8\subseteq A$,

(ii) $\frac A{32} \subseteq \frac A8$ because $\frac A4 \subseteq A$, and

(iii) $\frac A4 - \frac A8\subseteq A$.

Using the above we get $\frac A{32} - \frac A{32} \subseteq \frac A4 - \frac A8 \subseteq A$. Thus the set $B=\frac A{32} - \frac A{32}$ works. Clearly it is symmetric, and it is open and connected: It is connected since it is the continuous image of $A\times A$ under the subtraction function (and division by $32$), and the product space $A\times A$ is connected since the factors are.

Here are some details on the condition that $\frac A4\subseteq A$, which was used in the above proof. One way to prove it is without a reference to the closure $C$ of $A$, as follows.

Claim. $\frac A4 \subseteq A$ (or equivalently, $2A\subseteq8A$).

Proof. Take any $a\in A$, we need to show that $2a\in 8A$. Since $A$ is open, there is $n$ such that $a+\frac a{2\cdot8^n}\in A$. Then $2a=2(a+\frac a{2\cdot8^n})-\frac a{8^n}\in 2A-A\subseteq 8A$. This completes the proof of the Claim.

Here is an alternative way to show that $2A\subseteq8A$. Show that $2A-A=2A-C$ (where $A$ is open and $C$ is the closure of $A$). (Then, since $0\in C$ we get that $2A\subseteq2A-C=2A-A\subseteq8A$.)

Lemma. If $U$ is open and $K$ is arbitrary then $U+\overline K= U+K$ (where $\overline K$ is the closure of $K$).

Proof. Pick any $p\in U+\overline K$. Then $p=q+r$ for some $q\in U$ and $r\in\overline K$. Since $U$ is open, there exists a symmetric neighborhood $V$ of $0$ such that $q+V\subseteq U$. Pick $s\in(r+V)\cap K$. Then $v=s-r\in V$ so $-v\in-V=V$ and $q-v\in U$, hence $p=q+r=q-v+r+v=(q-v)+s\in U+K$. Thus $U+\overline K\subseteq U+K$ and $U+\overline K=U+K$.

Corollary. If $A$ is open and $C$ is the closure of $A$ then $2A-C=2A-A$. (So, if, in addition, $2A-A\subseteq8A$ where $A$ is open and nonempty then $2A\subseteq2A-C=2A-A\subseteq8A$, using that $0\in C$ for the first inclusion.)

Proof. Use the above Lemma with $U=2A$ and $K=-A$.

Discussion. So the inclusion $2A\subseteq8A$ was given two different proofs, one direct, and another using the closure $C$ of $A$, along with the above lemma and corollary (providing a quicker approach, at least to me). I did not know (beforehand) if $0\in A$, and did not use it in my proof (though it eventually follows from $0\in\frac A{32}-\frac A{32}=B\subseteq A$), and I do not know if $A\cap(-A)$ must be connected (assuming that $A$ is open and $2A-A\subseteq8A$). An example when $A\cap(-A)$ need not be connected was provided by another user in the comments above, $A$ is the union of the two closed upper semicircles in the plane, of radius $1$ with centers at $(\pm1,0)$, but this $A$ is not open and $2A-A\not\subseteq8A$. (The two semicircles could easily be made open, by ``thickening'' them a little, but it is not clear to me at present if we could also get $2A-A\subseteq8A$, and yet to have $A\cap(-A)$ disconnected). One more comment: If $0\in A$ (where $A$ is open) and if we work in a locally connected space, then the connected component of $A\cap(-A)$ containing $0$ would be open and symmetric (so this component could play the role of $B$). But, we need to assume some extra condition (e.g. $2A-A\subseteq8A$) to show that $0\in A$, and even if we knew that $0\in A$, it might not be immediately clear how $A\cap(-A)$ would possibly help, if the space is not locally connected. (I would be curious to see a proof - if there is one - based on the use of $A\cap(-A)$, showing that either it is connected, or that it contains a connected, open symmetric set.)

$\endgroup$
  • $\begingroup$ About your question in brackets, the answer is positive, see Theorem 1.15.a in Rudin's "Functional Analysis": let $V$ be any neighborhood of $0$, then $\bigcup_n nV=X$. $\endgroup$ – Paolo Leonetti May 15 at 14:46
  • $\begingroup$ @PaoloLeonetti I think I fixed the details in my proof (with or without the use of the closure $C$ of $A$). So now I know $0\in A$ since $0\in B\subseteq A$. I do not know if $A\cap(-A)$ is connected (and I am curious, but will leave it for now). Clearly $A\cup(-A)$ is connected. $\endgroup$ – Mirko May 15 at 15:29
  • $\begingroup$ The proof depends on the choice of the coefficients, however I find it very nice. $\endgroup$ – Paolo Leonetti May 16 at 14:12

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.