This is a follow-up on the following question. Let $\text{End}(X)$ denote the endomorphism monoid of a topological space $X$ (that is, the collection of all continuous maps $f:X\to X$ with composition).

What is an example of a topological space $X$ with $X\not\cong \mathbb{R}$ but the monoids $\text{End}(X)$ and $\text{End}(\mathbb{R})$ are isomorphic?

  • 13
    $\begingroup$ A general remark: in $\mathrm{End}(X)$, constants are precisely the left-absorbing elements (i.e., those $g$ such that $gf=g$ for all $f$). The monoid acts on the left on left-absorbing elements, and this is conjugate to the action on $X$. In particular, the condition that $X$ is homogeneous under $\mathrm{Aut}(X)$ can be read on the abstract monoid $\mathrm{End}(X)$. I expect this is might be a starting point to check that every $X$ with $\mathrm{End}(X)\simeq\mathrm{End}(\mathbf{R})$ is homeomorphic to $\mathbf{R}$. $\endgroup$ – YCor Mar 29 at 15:17
  • 9
    $\begingroup$ The constant maps are not enough in general to pick up the topology for a general space. For example the endomorphism monoid of X in the discrete and indiscrete topology are the same. $\endgroup$ – Benjamin Steinberg Mar 29 at 16:55
  • 1
    $\begingroup$ Indeed it sounds tricky, even assuming beforehand that $X$ is Hausdorff. $\endgroup$ – YCor Mar 29 at 17:57

No such space exists. We actually get the stronger statement that every isomorphism $\operatorname{End}(X) \stackrel\sim\to \operatorname{End}(\mathbf R)$ is induced by an isomorphism $X \stackrel\sim\to \mathbf R$ (unique by Observation 1 below). In contrast, in Emil Jeřábek's beautiful construction in this parallel post there is an 'outer automorphism' $\operatorname{End}(X) \stackrel\sim\to \operatorname{End}(X)$ that does not come from an automorphism $X \stackrel\sim\to X$ of topological spaces (it comes from an anti-automorphism of ordered sets).

I will use the substantial progress by YCor and Johannes Hahn, summarised as follows:

Observation 1 (YCor). For every topological space $X$, the map $X \to \operatorname{End}(X)$ taking $x$ to the constant function $f_x$ with value $x$ identifies $X$ with the set of left absorbing¹ elements of $\operatorname{End}(X)$.

In particular, an isomorphism of monoids $\operatorname{End}(X) \stackrel\sim\to \operatorname{End}(Y)$ induces a bijection $U(X) \stackrel\sim\to U(Y)$ on the underlying sets.

Observation 2 (Johannes Hahn). If $\operatorname{End}(X) \cong \operatorname{End}(\mathbf R)$, then $X$ is $T_1$. Since the closed subsets of $\mathbf R$ are exactly the sets of the form $f^{-1}(x)$ for $x \in \mathbf R$, we conclude that these are closed in $X$ as well, so the bijection $X \to \mathbf R$ of Observation 1 is continuous.

(The asymmetry is because we used specific knowledge about $\mathbf R$ that we do not have about $X$.)

To conclude, we prove the following lemma.

Lemma. Let $\mathcal T$ be the standard topology on $\mathbf R$, and let $\mathcal T' \supsetneq \mathcal T$ be a strictly finer topology. If all continuous maps $f \colon \mathbf R \to \mathbf R$ for $\mathcal T$ are continuous for $\mathcal T'$, then $\mathcal T'$ is the discrete topology.

Note that Observation 2 and the assumption $\operatorname{End}(X) \cong \operatorname{End}(\mathbf R)$ imply the hypotheses of the lemma, so we conclude that either $X = \mathbf R$ or $X = \mathbf R^{\operatorname{disc}}$. The latter is clearly impossible as it has many more continuous self-maps.

Proof of Lemma. Let $U \subseteq \mathbf R$ be an open set for $\mathcal T'$ which is not open for $\mathcal T$. Then there exists a point $x \in U$ such that for all $n \in \mathbf N$ there exists $x_n \in \mathbf R$ with $|x - x_n| \leq 2^{-n}$ and $x_n \not\in U$. Without loss of generality, infinitely many $x_n$ are greater than $x$, and we can throw out the ones that aren't (shifting all the labels, so that $x_0 > x_1 > \ldots > x$). Up to an automorphism of $\mathbf R$, we can assume $x = 0$ and $x_n = 2^{-n}$ for all $n \in \mathbf N$. Taking the union of $U$ with the usual opens $(-\infty,0)$, $(1,\infty)$, and $(2^{-n},2^{-n+1})$ for all $n \in \mathbf N$ shows that $$Z = \big\{1,\tfrac{1}{2},\tfrac{1}{4},\ldots\big\}$$ is closed for $\mathcal T'$. Consider the continuous function \begin{align*} f \colon \mathbf R &\to \mathbf R\\ x &\mapsto \begin{cases}0, & x \leq 0,\\ x, & x \geq 1, \\ 2^nx, & x \in \big(2^{-2n},2^{-2n+1}\big], \\ 2^{-n}, & x \in \big(2^{-2n-1},2^{-2n}\big].\end{cases} \end{align*} Then $f^{-1}(Z)$ is the countable union of closed intervals $$Z' = \bigcup_{n \in \mathbf N} \big[2^{-2n-1},2^{-2n}\big] = \big[\tfrac{1}{2},1\big] \cup \big[\tfrac{1}{8},\tfrac{1}{4}\big] \cup \ldots.$$ By the assumption of the lemma, both $Z'$ and $2Z'$ are closed in $\mathcal T'$, hence so is the union $$Z'' = Z' \cup 2Z' \cup [2,\infty) = (0,\infty),$$ and finally so is $Z'' \cup (-Z'') = \mathbf R\setminus 0$. Thus $0$ is open in $\mathcal T'$, hence so is every point, so $\mathcal T'$ is the discrete topology. $\square$

¹Elements $f$ such that $fg = f$ for all $g$. (I would probably have called this right absorbing!)

| cite | improve this answer | |
  • 3
    $\begingroup$ I checked before posting: the condition that $f$ satisfies $\forall g,fg=g$ is always called "left-absorbing", or "left zero" (which sounds intuitive to me: $f$ absorbs on the lefts elements from the right). $\endgroup$ – YCor Mar 30 at 14:34
  • 2
    $\begingroup$ @YCor: sorry, I wasn't trying to challenge your use. I also checked and saw that this is what it's called. Anyway, left and right are hard. $\endgroup$ – R. van Dobben de Bruyn Mar 30 at 22:12
  • 3
    $\begingroup$ No problem. In the same vein, in a group $G$ with subgroup $H$, one calls $gH$ left coset and $Hg$ right coset but this is possibly not obvious either. $\endgroup$ – YCor Mar 30 at 22:23
  • 1
    $\begingroup$ I feel that YCor's guys are "left absorbing" or "absorbing right" (in that order). $\endgroup$ – Wlod AA Mar 31 at 6:40
  • 1
    $\begingroup$ Over in the other thread I said "left-absorbing", but it means of course absorbing to the left! $\endgroup$ – Todd Trimble Mar 31 at 10:18

By @YCor's comment, $X$ has the same number of elements as $\mathbb{R}$ and the actions of $End(X)$ on $X$ is the same as the action of $End(\mathbb{R})$ on $\mathbb{R}$. Now consider the automorphism group and its action on $X$. $Aut(\mathbb{R})$ consists of strictly increasing and strictly decreasing maps and it is easy to see that the point stabiliser subgroups $Aut(\mathbb{R})_x$ act transitively on $\mathbb{R}\setminus\{x\}$. But this action is imprimitive: $\mathbb{R}\setminus\{x\}$ has two blocks, namely $(x,+\infty)$ and $(-\infty,x)$ and the set stabilisers act transitively on those two sets. Phrased differently: There are exactly three equivalence relations on $\mathbb{R}\setminus\{x\}$ invariant under $Aut(\mathbb{R})_x$, the two trivial ones "everything is equivalent" and "nothing is equivalent" as well as a unique non-trivial one with the two equivalence classes $(-\infty,x)$ and $(x,+\infty)$.

Conclusion: We can recover a linear order on $X$ from $Aut(X)$ and thus from $End(X)$. And in particular we can recover the order topology of this ordering. And all elements of $End(X)$ must be continuous w.r.t. this order topology.

This does not necessarily mean that the order topology coincides with the original topology on $X$, but it is very close.

What can we say about the original topology? We know that there are lots of continuous map, but not too many (since $|End(\mathbb{R})|=|\mathbb{R}|$) so that $X$ is not indiscrete. We can say that the only self-maps with finite image are constant. In particular, there cannot be a Sierpinski space inside $X$, because every open set yields a continuous map to the Sierpinski space. Therefore $X$ is at least a $T_1$-space.

That is enough to conclude that the original topology on $X$ is at least as fine as the order topology: For every order-topology-closed subset $A\subseteq\mathbb{R}$ there is a continuous self-map $f:\mathbb{R}\to\mathbb{R}$ with $A=f^{-1}(0)$. Therefore the corresponding order-topology-closed subset of $X$ is also the preimage of a point under a continuous self-map and thus original-topology-closed. In other words: Our canonical bijection $X\to\mathbb{R}$ is continuous w.r.t. the original topology on $X$.

I have the feeling that the other direction is equally easy, but I just don't see it right now.

| cite | improve this answer | |
  • 3
    $\begingroup$ If you have a continuous bijection onto $\mathbf{R}$, you have the Hausdorff property. I've been wondering if there are strange strong topologies (such as the one generated by the standard topology and the co-countable subsets), maybe the latter has the same homeomorphism group, but not the same monoid of continuous self-maps). $\endgroup$ – YCor Mar 29 at 18:26
  • $\begingroup$ Thanks for this effort that is a key ingredient towards answering the question! $\endgroup$ – Dominic van der Zypen Mar 30 at 8:45

There is a straightforward approach to proving that isomorphism of monoids $\ \text{End}(X\ T)\ $ and $\ \text{End}(\Bbb R\,\ T_E)\ $ implies homeomorphism of $\ (X\ T)\ $ and Euclidean space $\ (\Bbb R\,\ T_E):\ $

let $\ (M\ \circ\ J)\ $ be an arbitrary abstract monoid. One may associate with this monoid a set of "points" $\ C\ $ of "constant" elements, as @YCor did in his first comment, as the left-absorbing elements $\ c\in C,\ $ where

$$ \forall_{f\in M}\quad c\circ f=c $$

Then one selects purely algebraic (monoidal) notions which have the respective topological meaning in $\ (\Bbb R\,\ T_E).\ $ You need just a couple of such notions.

Then you force on the abstract monoid $\ (M\ \circ\ J)\ $ the purely algebraic (monoidal) axioms that induce the properties of reals. That's all.

For instance one may apply idempotents $\ i\in\mathcal I\subseteq M,\ $ where $\ i\circ i=i. \ $ Then the critical notion to me is what I've defined and call universally closed idempotents or uc-morphism $\ i\in\mathcal I,\ $ which satisfy:

$$ \forall_{f\in M}\,\exists_{p\in M}\quad i\circ f\circ i\circ p\,\ =\,\ i\circ p $$

Observe that we have a canonical map

$$ \pi: \mathcal I\to 2^C, $$


$$ \forall_{i\in\mathcal I}\quad \pi(i)\ := \ \{c\in C:\ i\circ c=c\} $$

This has nice properties... etc.


(After I saw the OP question, I hesitated... and I can remove my post, no sweat.)

| cite | improve this answer | |
  • 1
    $\begingroup$ Also, idempotents that are not universally closed (or nuc-morphisms for short) have their role as well. $\endgroup$ – Wlod AA Mar 31 at 6:47
  • $\begingroup$ Dominic shoots from the hip on MO but this OP-question and his other one, EndMonDom-question are well aimed! The latter offers great scope and allows for a general presentation of the topic. (I saw this latter but earlier question only now). $\endgroup$ – Wlod AA Mar 31 at 14:35

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.