I know that this is a well known result, but where can I find a proof? I am also interested to see more general non-embedding results of this type.

Theorem. Let $Y$ be the union of two segments with one segment being attached in the middle of the other one. Then there is no topological embedding of the space $Y\times Y$ into $\mathbb{R}^3$.

It was a homework problem in my undergraduate (sophomore) topology class taught by Professor Karol Sieklucki at University of Warsaw!

  • 1
    $\begingroup$ @AntonPetrunin This was a standard course of topology of metric spaces, but there were 4 IMO students in the class and the problems were far beyond of what was covered in lectures. Another problem was the Moore-Young theorem: mathoverflow.net/a/27248/121665 $\endgroup$ Apr 13, 2018 at 17:43
  • 1
    $\begingroup$ Okey, did you get to homology groups (maybe I am wrong, but it looks like a senseless pain without them). $\endgroup$ Apr 13, 2018 at 17:58
  • 4
    $\begingroup$ No idea. For me this is a folklore result and I do not know any references. This is why I ask. $\endgroup$ Apr 13, 2018 at 20:50
  • 1
    $\begingroup$ Piotr, the question can be essentially equivalently asked about embeddings into $S^3$. Years ago, Professor Karol Borsuk asked about the non-embeddability of nice products $\ X\times Y$ into $S^n$. As I remember, He got the first result for starters, then others followed (not too many of them), including one or both parents of Greg K., then I presented a strong result during a conference in Auburn (AL) but didn't publish it; then someone from Poland published a similar but weaker result -- that's all I remember at this moment; I will not recall much extra later after all these years. $\endgroup$
    – Wlod AA
    Jan 17, 2019 at 4:22
  • 1
    $\begingroup$ "two segments with one segment being attached in the middle of the other one" is not the most elegant way to describe this space. (:-) $\endgroup$
    – Wlod AA
    Jan 17, 2019 at 4:26

3 Answers 3


If it did embed as $Y \times Y \subset \mathbb{R}^3$, then you could choose a sufficiently small sphere $S$ in $\mathbb{R}^3$ centered at the point $(y_0,y_0)$, where $y_0$ is the point in where the two segments are glued. Then $S \cap Y \times Y \subset S$ is an embedding of the complete bipartite graph $K_{3,3}$ into a two sphere. But this graph is not planar, which is a contradiction. This argument seems to assume the embedding is sufficiently nice, e.g. simplicial with respect to some triangulations of $Y \times Y$ and $\mathbb{R}^3$. Maybe others can share a different argument that is valid for any topological embedding.

  • 2
    $\begingroup$ Could you elaborate why the embedding is $K_{3,3}$. I don't quite "see" this. $\endgroup$
    – Michael
    Jan 17, 2019 at 1:31
  • $\begingroup$ Think of $Y$ as consisting of three edges $E, F, G$ glued at a vertex $v$. A small sphere in $Y$ centered at $v$ consists of three points, one for each of $E,F,G$. A small sphere in $Y \times E$ centered at $(v,v)$ consists of four points, $E,F,G,E'$, where $E'$ lies in $\{v\} \times E$. All of $E,F,G$ are joined to $E'$ by arcs, one for each of the squares $E \times E$, $F \times E$, $G \times E$. The same is true in $Y \times F$ and $Y \times G$. Now glue these three spaces together (along $Y \times \{v\}$) to form $Y \times Y$. The spheres fit together to form $K_{3,3}$. $\endgroup$ Jan 17, 2019 at 12:51

Theorem 10 of "An Alternative Proof that 3-Manifolds Can be Triangulated" by R.H. Bing [Ann. of Math. Vol. 69, (1959), pp. 37-65] states that any topologically embedded simplicial 2-complex $K$ in a triangulated 3-manifold $M$ is near a PL embedding of $K$ into $M$.

In particular, a $2$-dimensional finite simplicial complex has a topological embedding in $\mathbb R^3$ if and only if it has a PL embedding in $\mathbb R^3$. Together with Robert Bell's answer this resolves the question.


I will show that $Y^k$ cannot embed in $\mathbb{R}^p$ when $p < 2k$.

Write $\mathrm{Conf}(2,X) = \{(x,x') \in X^2 \, | \; x \neq x'\}$ for the deleted diagonal of $X$, and note that any injection $Y \hookrightarrow X$ induces an $S_2$-equivariant map $\mathrm{Conf}(2,Y) \to \mathrm{Conf}(2, X)$.

We need the following corollary of Theorem 1.8 from my paper "Configuration space in a product".

If an embedding $A \subseteq B$ induces a homotopy equivalence of pairs $(A^2, \mathrm{Conf}(2,A)) \simeq (B^2, \mathrm{Conf}(2,B))$, then it also induces a homotopy equivalence $\mathrm{Conf}(2,A^k) \simeq \mathrm{Conf}(2,B^k)$ for all $k \geq 0$.

Here is a short proof of the corollary in the case $k=2$.

Two points in $A^2$ are distinct if and only if they are distinct in their first coordinate, or their second coordinate, or both. In other words, $\mathrm{Conf}(2,A^2)$ is covered by the two open subsets $A \times \mathrm{Conf}(2,A)$ and $\mathrm{Conf}(2,A) \times A$, and the intersection of these sets is $\mathrm{Conf}(2,A) \times \mathrm{Conf}(2,A)$. As a consequence, $\mathrm{Conf}(2,A^2)$ is the homotopy pushout of a diagram that depends only on the pair $(A^2, \mathrm{Conf}(2,A))$. By our assumption, the inclusion $A \subseteq B$ induces a pointwise homotopy equivalence on these pushout diagrams, and hence on homotopy pushouts. This concludes the proof for $k=2$; the case of general $k$ requires a larger homotopy colimit, but is otherwise similar.

"Reordering the cars in the driveway twice" gives an equivalence $\mathrm{Conf}(2,Y) \simeq \mathrm{Conf}(2, \mathbb{R}^2)$, (even when restricted to the driveway and a little part of the street, two cars may wind around each other), leading to an equivalence of pairs $$ (Y^2, \mathrm{Conf}(2,Y)) \simeq (\mathbb{R}^4, \mathrm{Conf}(2,\mathbb{R}^2)) $$ induced by the usual inclusion $Y \subset \mathbb{R}^2$. By the corollary, $\mathrm{Conf}(2, Y^k)$ is homotopy equivalent to $\mathrm{Conf}(2, \mathbb{R}^{2k})$, and moreover, this map is $S_2$ equivariant.

Since $\mathrm{Conf}(2, \mathbb{R}^{p}) \simeq_{S_2} (S^{p-1}, \tau)$, where $\tau$ denotes the antipodal action, and similarly $\mathrm{Conf}(2,Y) \simeq_{S_2} \mathrm{Conf}(2, \mathbb{R}^{2k}) \simeq_{S_2} (S^{2k-1}, \tau)$, any embedding $Y^{k} \subseteq \mathbb{R}^p$ induces an $S_2$-map $$ (S^{2k-1}, \tau) \to (S^{p-1}, \tau), $$ which is impossible for $p<2k$ by the Borsuk-Ulam theorem.

  • $\begingroup$ Thank you for your proof. I will read it carefully, but I need time (busy). $\endgroup$ Jan 30, 2019 at 17:53

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.