I am new studying additive subgroups of the real line, I would like to know if someone could give me an idea for the next question.

Let $m$ be the Lebesgue measure in $\mathbb{R}$. A measurable set $E\subseteq\mathbb{R}$ has density $d$ at $x$ if $$\lim_{h\to 0} \frac{m(E\cap [x-h, x+h])}{2h} $$ exists and equals $d$. Denote by $\phi(E)$, $\{x\in\mathbb{R} : d(x, E)=1\}$.

The family of all measurable sets $E$ such that $E\subseteq\phi(E)$ is a topology on $\mathbb{R}$, henceforth denoted by $(X, \mathcal{T})$ or just X if confusion is unlikely. Clearly $\mathcal{T}$ is stronger that the usual topology $(\mathbb{R}, \mathcal{E})$, that is, $\mathcal{E}\subseteq\mathcal{T}$. This topology is called the density topology in $\mathbb{R}$.

Some properties of the density topology.


  • The Borel subsets of $X$ are precisely the measurable sets.
  • Every Borel subset of $X$ is a $G_{\delta}$.
  • Every regular open set is a Euclidean $F_{\sigma \delta}$.
  • $X$ satisfies the countable chain condition.
  • $X$ is neither separable nor first countable, but every subspace of $X$ is Baire.


The following conditions on a subset $Y$ of $X$ are equivalent:

  • $Y$ is a nullset (i.e. has measure zero)
  • $Y$ is a nowhere dense
  • $Y$ is a first category
  • $Y$ is closed discrete.

My question is the following :

Suppose $G$ is an additive subgroup of $\mathbb{R}$ of positive Lebesgue outer measure such that $G$ is of the first category in $(\mathbb{R}, \mathcal{E})$. How can I conclude that $G$ is dense in $(\mathbb{R}, \mathcal{T})$?

Remember that the Lebesgue inner measure of $E\subseteq \mathbb{R}$ is defined as
$$m_{*}(E)=\sup\{m(C) : C\subseteq E, C\hspace{0.1cm} \text{is}\hspace{0.1cm}\mathcal{E}-\text{closed} \} $$

In general, we have the following characterization for dense subsets in the density topology on $\mathbb{R}$.

Theorem. A subset $D$ of $\mathbb{R}$ is $\mathcal{T}$-dense in $\mathbb{R}$ iff $m_{*}(\mathbb{R}\setminus D)=0$.

Proof. Suppose that $D$ is $\mathcal{T}$-dense in $\mathbb{R}$, then $\text{int}_{\mathcal{T}}(\mathbb{R}\setminus D)=\emptyset$. Let $C$ be a closed set of $(\mathbb{R}, \mathcal{E})$ such that $C\subseteq \mathbb{R}\setminus D$, in particular $C$ is $\mathcal{T}$-closed, then $\text{int}_{\mathcal{T}}(\overline{C}^{\mathcal{T}})=\text{int}_{\mathcal{T}}(C)\subseteq \text{int}_{\mathcal{T}}(\mathbb{R}\setminus D)=\emptyset$, by FACT 2, $m(C)=0$, then $m_{*}(\mathbb{R}\setminus D)=0$.

Now, suppose that $D$ is not $\mathcal{T}$-dense, then there is $A\in \mathcal{T}\setminus \{\emptyset \}$ such that $A\cap D=\emptyset$, so $A\subseteq \mathbb{R}\setminus D$, therefore $m(A)=0$, contradiction (because every non-empty $\mathcal{T}$-open subset of $\mathbb{R}$ has positive measure).


2 Answers 2


I think that below I manage to answer the first of your questions. I will be very grateful for verification of this argument. Unless it is correct, I will delete this answer.

I denote Lebesgue outer measure by $m^*$ and $\sigma$-algebra of Lebesgue measurable sets by $\mathcal{L}$. Let $\mathcal{B}(\mathbb{R})$ be the $\sigma$-algebra of Borel sets on $\mathbb{R}$ with respect to the usual topology. Suppose that $G$ is an additive subgroup of $\mathbb{R}$ such that $m^*(G)>0$.

Lemma 1. $G$ is dense in $\mathbb{R}$ with respect to the usual topology.

Proof. In the proof of this lemma we consider $\mathbb{R}$ with the usual topology.

Suppose that for every $\epsilon >0$ there exists $g\in G$ such that $g\in (-\epsilon,\epsilon)\setminus \{0\}$. Then for every $x\in \mathbb{R}$ there exists $n\in \mathbb{Z}$ such that $|n\cdot g - x|<\epsilon$. Thus $G$ is dense in $\mathbb{R}$.

Now if $G$ is not dense in $\mathbb{R}$, then by what we said above, there must be $\epsilon>0$ such that $\{0\} = G\cap (-\epsilon, \epsilon)$. This means that $0$ is isolated in $G$. Since $G$ is a topological group, it is homogeneous (any two points have homeomorphic neighborhoods - just pick translation). We deduce that every point of $G$ is isolated and hence $G$ is a discrete subset of $\mathbb{R}$. A discrete subgroup of $\mathbb{R}$ is countable and hence $m^*(G)=0$. This is contradiction with $m^*(G)>0$.

Lemma 2. There exists a constant $c>0$ depending only on $G$ such that for every bounded interval $I\subseteq \mathbb{R}$ we have $$m^*\left(G\cap I\right) = c\cdot m^*(I)$$ Proof. Fix $a\in \mathbb{R}$ and consider a function $f_a:(a,+\infty)\rightarrow \mathbb{R}$ given by formula $$f_a(x) = m^*\left((a,x]\cap G\right)$$ Next for any $h>0$ pick a function $D_{a,h}:(a,+\infty)\rightarrow \mathbb{R}$ given by formula $$D_{a,h}(x)= f_a(x+h)-f_a(x) =m^*\left((a,x+h]\cap G\right) - m^*\left((a,x]\cap G\right) =$$ $$= m^*\left((x,x+h]\cap G\right)$$ The last equality follows from the Caratheodory's criterion. Since $G$ is an additive subgroup of $\mathbb{R}$ and Lebesgue outer measure is translation invariant, we derive that

$$D_{a,h}(x+g) = D_{a,h}(x)$$

for every $g\in G$ and $g>0$. By Lemma 1 we know that $G_+ = \{g>0|g\in G\}$ is dense in $\mathbb{R}_+$. Now the fact that $D_{a,h}$ is continuous implies that

$$D_{a,h}(x+t) = D_{a,h}(x)$$

for every $t>0$. Thus $D_{a,h}$ is constant for every $h>0$ and this implies that monotone function $f_a$, which has derivative almost everywhere, has constant derivative everywhere. Say $f_a'(x) = c$ for every $x\in (a,+\infty)$. Next $c$ does not depend on $a<0$, because for $a_1,a_2$ functions $f_{a_1}$ and $f_{a_2}$ have the same slope on $\left(\max\{a_1,a_2\},+\infty\right)$. Moreover, $c>0$, since $m^*(G)>0$.

Now for given $a\in \mathbb{R}$ we have $f_a(a) = 0$ and $f_a'(x) = c$ for $x\in (a,+\infty)$. This implies that
$$m^*\left(G\cap (a,x]\right) = f_a(x) = c\cdot x - c\cdot a$$ for every $a\in \mathbb{R}$. If $I = (\xi_1,\xi_2]$ is an interval, then pick $a< \xi_1$ and then

$$m^*(G\cap I) = m^*(G\cap (\xi_1,\xi_2]) = m^*(G\cap (a,\xi_2]) - m^*(G\cap (a,\xi_1]) =$$ $$= f_a(\xi_2) - f_a(\xi_1) = \left(c\cdot \xi_2 - c\cdot a\right) - \left(c\cdot \xi_1 - c\cdot a\right) = c\cdot (\xi_2 - \xi_1) = c\cdot m^*(I) $$

Thus we proved that the statement of the lemma holds for intervals open from the left and closed from the right. Since every interval is up to (endpoints) a set of measure zero of the above form, we deduce that the result holds for all intervals.

Lemma 3. Let $E$ be a subset of $\mathbb{R}$ and let $c>0$. Suppose that $m^*(E\cap I) = c\cdot m^*(I)$ for every bounded interval $I\subseteq \mathbb{R}$. Then for every bounded set $A\in \mathcal{L}$ we have $$m^*(E\cap A) = c\cdot m^*(A)$$

Sublemma. Let $\{A_n\}_{n\in \mathbb{N}}$ be a family of pairwise disjoint members of $\mathcal{L}$ and let $E\subseteq \mathbb{R}$ be a bounded subset. Then $$m^*\left(E\cap \bigcup_{n\in \mathbb{N}}A_n\right) = \sum_{n\in \mathbb{N}}m^*(E\cap A_n)$$ Proof of the sublemma. It is a consequence of the fact that $\mathcal{L}$ is constructed via Caratheodory's criterion that this result holds for finite family $\{A_n\}_{n=0}^N$ of pairwise disjoint sets in $\mathcal{L}$. We use this in the proof of the general case. We have $$m^*(E) \leq m^*\left(E\cap \bigcup_{n\in \mathbb{N}}A_n\right) + m^*\left(E\setminus \bigcup_{n\in \mathbb{N}}A_n\right)\leq \sum_{n\in \mathbb{N}}m^*(E\cap A_n) + m^*\left(E\setminus \bigcup_{n\in \mathbb{N}}A_n\right) =$$ $$\leq \lim_{N\rightarrow +\infty}\left(\sum_{n=0}^Nm^*(E\cap A_n) + m^*\left(E\setminus \bigcup_{n=0}^NA_n\right)\right) =$$ $$ = \lim_{N\rightarrow +\infty}\left(m^*\left(E\cap \bigcup_{n=0}^NA_n\right) + m^*\left(E\setminus \bigcup_{n=0}^NA_n\right)\right)=\lim_{N\rightarrow +\infty}\mu^*(E) =\mu^*(E)$$ Thus in the inequality above, we have equality everywhere. In particular, we have $$ m^*\left(E\cap \bigcup_{n\in \mathbb{N}}A_n\right) + m^*\left(E\setminus \bigcup_{n\in \mathbb{N}}A_n\right)= \sum_{n\in \mathbb{N}}m^*(E\cap A_n) + m^*\left(E\setminus \bigcup_{n\in \mathbb{N}}A_n\right)$$ The boundness of $E$ implies that we can cancel out $m^*\left(E\setminus \bigcup_{n\in \mathbb{N}}A_n\right) < +\infty$.

Proof of the lemma 3. We prove that for every bounded interval $I$ and $A\in \mathcal{B}(\mathbb{R})$ such that $A\subseteq I$ we have

$$m^*(E\cap A) = c\cdot m^*(A)$$

Note that the family $\mathcal{F}$ of all such $A$ is a Dynkin system in the power set $\mathcal{P}(I)$. Indeed, if $A\in \mathcal{F}$ that is $m^*(E\cap A)= c\cdot m^*(A)$, then $$c\cdot m^*(I) = m^*(E\cap I) = m^*(E\cap A) + m^*(E\cap (I\setminus A)) =$$ $$= c\cdot m^*(A) + m^*(E\cap (I\setminus A)) $$ Hence $m^*(E\cap (I\setminus A)) = c\cdot m^*(I\setminus A)$ and thus $I\setminus A\in \mathcal{F}$. Moreover, from Sublemma we derive that $\mathcal{F}$ is closed under countable unions of pairwise disjoint sets. By the assumption $\mathcal{F}$ contains all subintervals of $I$ and they form $\pi$-system. Now by Dynkin's $\pi\lambda$-theorem $\mathcal{F}$ contains all subsets in $\mathcal{B}(\mathbb{R})$ which are contained in $I$. Suppose that $A\in \mathcal{L}$ is contained in $I$, then $A = B\cup Z$, where $B\subseteq I$ and $B\in \mathcal{B}(\mathbb{R})$ and $m^*(Z)=0$. Hence

$$m^*(E\cap B) \leq m^*(E\cap A) = m^*\left((E\cap B)\cup (E\cap Z)\right) \leq m^*(E\cap B) + m^*(E\cap Z) = m^*(E\cap B) + 0 = m^*(E\cap B)$$

Thus $m^*(E\cap B) = m^*(E\cap A)$ and

$$m^*(E\cap A) = m^*(E\cap B) = c\cdot m^*(B) = c\cdot m(B) = c\cdot m(A) = c\cdot m^*(A)$$

This proves the lemma.

Now by Lemma 2 and Lemma 3, we derive that there exists $c>0$ such that $m^*(G\cap A) = c\cdot m^*(A)$ for every bounded measurable subset $A$ of $\mathbb{R}$. Now pick any nonempty subset $A\in \mathcal{T}$. Intersect $A$ with some open interval $I$ such that $A\cap I\neq \emptyset$. Then $B = A\cap I\in \mathcal{T}$ is bounded. Thus $$m^*(G\cap A) = m^*(G\cap B)= c\cdot m^*(B)>0$$ because $c>0$ and $m^*(B)>0$. Thus $G\cap A\neq \emptyset$. Therefore, $G$ is dense in $\mathcal{T}$.

  • 1
    $\begingroup$ Thank you very much Slup for your response, I'll start studying it right away. $\endgroup$ Nov 16, 2019 at 4:58
  • $\begingroup$ @GabrielMedina I hope that this answer is correct, otherwise I am very sorry for wasting your time. Please let me know if I can help you with any clarifications. Thank you for your response. $\endgroup$
    – Slup
    Nov 17, 2019 at 8:55
  • $\begingroup$ Hello @Slup, I think they are very good ideas, for the proof of Lemma 1, please do you think you could be more explicit in the final part of the proof ? Another observation is that it seems to me that you do not need the hypothesis that $G$ is of first category in the usual topology. Thanks $\endgroup$ Nov 17, 2019 at 19:44
  • $\begingroup$ Hi @GabrielMedina I added more details to the proof of Lemma 1 and I hope now it is more clear. In the argument I used only the fact that $m^*(G)>0$ and eventually the argument shows that $m^*(G\cap A)>0$ for every Lebesgue set $A$ of positive measure, which is a very interesting property. $\endgroup$
    – Slup
    Nov 19, 2019 at 7:00
  • $\begingroup$ Thanks a lot @Slup. $\endgroup$ Nov 19, 2019 at 13:24

Remember the Steinhaus's Lemma,

Steinhaus's Lemma If $A\subseteq \mathbb{R}$, is a set of positive Lebesgue measure, then the set $A-A=\{x-y: x,y \in A\}$ contains a ball around $0$.

Corollary 1.1 If $S$ is an additive subgroup of $\mathbb{R}$, and $S$ contains a set of positive measure, then $S=\mathbb{R}$. If $T$ is a multiplicative subgroup of $]0, \infty[$ containing a set of positive measure then $T=]0, \infty[$.

Also, in the article "On two halves being two wholes" of Andrew Simoson, it is defined

Definition A subset $A$ of $\mathbb{R}$ is an Archimedean set if the set of all real numbers $r$ such that $A + r = A$ is dense in $\mathbb{R}$.

and it is shown that

Theorem 2 Let $A$ be an Archimedean set with positive outer measure. Then for any interval $I$, $$m^{*}(A \cap I) = m^{*}(I)$$

Proposition 2.1 Any additive subgroup of the additive group of real numbers is either cyclic (i.e., equal to $c\mathbb{Z}$ for some positive number $c$) or dense.

Corollary Let $G$ be an additive subgroup of the real line such that $m^{*}(G)>0$, then $G$ is dense.

Proof. Suppose that $G$ is not dense then, by Proposition 2.1, $G$ is cyclic, then $G$ is countable, and therefore $m^{*}(G)=0$, contradiction.

Proposition 3 Let $G$ be an additive subgroup of $\mathbb{R}$ such that $m^{*}(G)>0$. Then $G$ is dense in $(\mathbb{R}, \mathcal{T})$.

Proof. We start with the following

Lemma 3.1 Let $G$ be an additive subgroup of $\mathbb{R}$ such that $m^{*}(G)>0$, then for any interval $I$, $m^{*}(G\cap I)=m^{*}(I)$.

Proof of Lemma 3.1 Note that if $m^{*}(G)>0$, then $G$ is dense, so $G$ is an Archimedean set.

Finally, by the Lemma 3 of @Slup, we conclude the result.

  • 1
    $\begingroup$ (+1). It is nice that Theorem 2 is a stronger version of the lemma 2 from my answer. Can you please provide a link for the Andrew Simoson's paper? I have hard time looking for it. $\endgroup$
    – Slup
    Nov 26, 2019 at 7:57
  • $\begingroup$ Ok. I found it. $\endgroup$
    – Slup
    Nov 26, 2019 at 10:10
  • $\begingroup$ Here's an inline link to Article. $\endgroup$ Nov 26, 2019 at 12:41
  • $\begingroup$ Thanks. Btw Corollary 2.1 holds for every group of positive outer measure. Indeed, if the group is measurable and of positive outer measure, then it is $\mathbb{R}$ by Corollary 1.1 and the statement holds. Also in the Proposition 3 you don't need that your group is meager. $\endgroup$
    – Slup
    Nov 26, 2019 at 12:50
  • $\begingroup$ You're right, thanks for the observation. $\endgroup$ Nov 26, 2019 at 12:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.