Consider two topological spaces:

  1. Begin with $X_{>0}$ the positive dyadic and ternary rationals (including products e.g. $\frac n6$), with their topology as a subspace of $\Bbb R$. Then consider the quotient space $X_{>0}/\langle3\rangle$ given by completing the equivalence relation $x\sim 3x$ or in other words:

    $x\sim y\iff\exists i\in\Bbb Z:3^ix=y$

  2. As above, but now begin with $X_{>1}$ the dyadic and ternary rationals (including products) greater than $1$ and take the quotient $X_{>1}/\langle3\rangle$ using the same equivalence relation.


Do these quotient spaces have a nontrivial topology?


$\Bbb R$ is a metric space. Therefore the quotient topology is the same as the toplogy given by the quotient pseudometric, right? Or are there other Hausdorff quotient topologies?

Then for case 1: its topology IS trivial because I can pick arbitrarily large $i$ such that $d(3^{-i}x,3^{-i}y)$ is arbitrarily small in the vicinity of $0$.

For case 2: There's a prima facie argument that this quotient is Hausdorff because for any pair $x,y$ there's some maximal $i_1,i_2$ such that $\lvert 3^{i_1}x-3^{i_2}y\rvert$ is a difference between two representatives greater than $1$ - guaranteeing disjoint vicinities around them. But this isn't conclusive because correctly using the quotient metric involves finding the shortest distance between two elements in $X_{>1}/\langle3\rangle$ using any sequence of stepping stones in-between. I'm pretty much baffled how to do that. How do I determine a non-recursive function for the pseudometric?


For the reasons given in this answer: https://mathoverflow.net/a/380398/91341 which is largely the same as this problem, I am satisfied that the quotient pseudometric is the trivial $\forall x,y:d(x,y)=0$. I am of the view however that the quotient topology itself is NOT trivial, although I cannot show it.


1 Answer 1


Let $[x],[y]\in X_{>1}/\langle3\rangle$ with $x,y\in X_{>1}$ the least element of their respective equivalence classes. I claim that \begin{equation} \label{eq:pseudometric} d([x],[y])=|3^{-i}x-3^{-j}y| \tag{1} \end{equation} where $i$ and $j$ are the unique nonnegative intergers for which $1\leq 3^{-i}x,3^{-j}y<3$. Note that $3^{-i}x$ and $3^{-j}y$ are only contained in $[x]$ and $[y]$ if they are ternary. However, since the ternary rationals are dense in $\mathbb{R}$ it suffices to prove (1) for both $x$ and $y$ ternary. Accordingly, suppose both $x$ and $y$ are both ternary and the least element of their equivalence class. Thus, $1\leq x, y<3$ so that (1) asserts that $d([x],[y])=|x-y|$. If $a\sim x$ and $b\sim y$ with $a\neq x$ then $|a-b|>|x-y|$ so that $|x-y|<|x-a|+|a-b|+|b-y|$. Thus, the infimum of all the lengths of chains from $x$ to $y$ is $|x-y|$ so that (1) holds.

  • $\begingroup$ Apologies, I realise I have been ambiguous at best and misleading at worst in my question. Did you include e.g. $\frac76$ in $X$ which is a product of both dyadic and ternary rationals? $\endgroup$ Dec 24, 2020 at 8:54
  • 1
    $\begingroup$ Sorry, I did not realise you meant to include products. $\endgroup$ Dec 24, 2020 at 8:58
  • $\begingroup$ Please don't delete. As the case of ternary rationals can still serve as a useful example for me to think about, which I plan to do today... if not correctable I may also be able to edit the question to suit your answer. $\endgroup$ Dec 24, 2020 at 8:59
  • $\begingroup$ For each $x\in X_{>1}$ there exists a unique nonnegative integer $i$ for which $1\leq 3^{-i}x<3$. Note that $x^*=3^{-i}x\in [x]$ and is the least element. Let $[x],[y]\in X_{>1}/\langle3\rangle$. I claim that $$ d([x],[y])=|x^*-y^*|. \tag{1}$$ Let $\{a_0,a_1,\dots,a_n\}$ and $\{b_0,b_1,\dots,b_n\}$ be any chain between $x$ and $y$. Then, $$\sum_{i=0}^{n-1} |b_i-a_{i+1}| \geq \sum_{i=0}^{n-1} |b_i^*-a_{i+1}^*| =|x^*-a_1^*|+|b_1^*-a_1^*|+\cdots +|b_{n-1}^*-y^*|\geq |x^*-y^*|. $$ Hence, (1) holds. $\endgroup$ Dec 24, 2020 at 11:51
  • 1
    $\begingroup$ Apologies for the delay, I was unable to give some time to this for the past few days. Are we in agreement that $d([26],[10])=\left\lvert\dfrac{26}9-\dfrac{30}9\right\rvert=\dfrac49<\left\lvert\dfrac{26}9-\dfrac{10}9\right\rvert$ is a counterexample to your claimed proof? $\endgroup$ Jan 1, 2021 at 14:47

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.