I am struggling with the following question from Allen Hatcher's algebraic topology book.

Define the join $X*Y$ of two topological spaces $X$ and $Y$ to be the quotient of $X\times Y \times I$ under the following identifications:

$(x,y,0)\equiv (x,y',0)$ for all $y,y'$ in $Y$ $x\in X$

$(x,y,1)\equiv(x',y,1)$ for all $x,x'\in X$ and $y\in Y$

So $X*Y$ can be thought of as the union of all line segments joining points of $X$ to points of $Y$.

I am trying to prove that if $X$ is path connected and $Y$ is any space, then $X*Y$ is simply connected.

Here are two approaches I have tried:

I have read that the join $X*Y$ is homotopy equivalent to the suspension of the smash product, $\Sigma( X\wedge Y)$. So one possibility is to prove that the smash product is path-connected, and then to use the claim that the suspension of a path-connected space is simply-connected.

Another possibility is to show that the join is homeomorphic to the subspace $Cone(X)\times Y\cup X\times Cone(Y)$ of the space $Cone(X)\times Cone(Y)$. Then, if I could show that both sets in the above union are open (in the union) and simply-connected, and that their intersection is simply-connected, van Kampen's theorem would imply the result.

My problem with the first approach is that, as far as I know, the smash product depends on the choice of base points, so I am not even sure what the phrase "the join $X*Y$ is homotopy equivalent to the suspension of the smash product, $\Sigma( X\wedge Y)$" really means.

The second approach seems a bit more inviting, especially since the problem appears in Hatcher's book in the section of van Kampen's theorem; but I don't know how to prove that the sets$Cone(X)\times Y$ and $X\times Cone(Y)$ are open in their union, nor do I know how to show that they are simply-connected, unless both $X$ and $Y$ are simply-connected.

Any help (even for the case where $X$ and $Y$ are both simply connected) would be greatly appreciated.




3 Answers 3


Let $\{Y_i\}_i$ be the collection of path components of $Y$, so then $X*Y=\bigcup_i X*Y_i$. Let $Z$ be the portion of $X*Y$ associated with $[0,1/2)$, and let $A_i=Z\cup (X*Y_i)$. The intersection of any two of these $A_i$ is $Z$, which is path connected (deformation retracting onto $X$), so if we prove that $X*Y_i$ is simply connected, then $X*Y$ is simply connected by the van Kampen theorem. While it is true that the $A_i$ are not open in $X*Y$, this is fine because the surjectivity part of the theorem only relies on $f^{-1}(A_i)$ being open in $[0,1]$ for any path $f:[0,1]\to X*Y$. We defer showing openness to the appendix.

Because of this, let us assume $Y$ is path connected as well. The space $X*Y$ decomposes into two open subsets, one associated with $[0,2/3)$ and the other associated with $(1/3,1]$. Call these $Z_1$ and $Z_2$. The first deformation retracts onto $X$, the second deformation retracts onto $Y$, and their intersection $Z_1\cap Z_2$ deformation retracts onto $X\times Y$. The inclusion maps $X\times Y\to Z_1$ and $X\times Y\to Z_2$ induce projections on $\pi_1$, so the van Kampen theorem gives $\pi_1(X*Y)$ to be the free product $\pi_1(X)*\pi_1(Y)$ modulo the subgroup generated by all $i_{1*}(x,y)i_{2*}(x,y)^{-1}$, which are all $xy^{-1}$. Since $x,y$ may be arbitrary, the resulting $\pi_1(X*Y)$ is trivial, so $X*Y$ is simply connected.

Appendix. First, $Z$ is open in $X*Y$, so $f^{-1}(Z)$ is open. Take a point $f(s)=(x,y,t)$ in $X*Y_i$ with $t>0$. By continuity of $f$ there is a $\delta$ such that $f((s-\delta,s+\delta))\subset X\times Y\times (0,1]$. Since the image of this interval must lie in a path component, $f((s-\delta,s+\delta))\subset A_i$.

  • $\begingroup$ @EricWofsey That is completely true, but I didn't actually need anything beyond the ability to decompose paths into a composition of paths from the various $X*Y_i$. (This is Lemma 1.15 in Hatcher.) To do the decomposition using compactness of $[0,1]$, we just need to see that the inverse images of each piece of the decomposition are open, which I hopefully didn't make a mistake showing. $\endgroup$ Oct 20, 2016 at 8:37
  • $\begingroup$ Yep, that argument looks good now. $\endgroup$ Oct 20, 2016 at 17:55
  • $\begingroup$ Oh, except that you should also mention that the inclusion $X*Y_i\to A_i$ is a homotopy equivalence. This again takes a little bit of care to prove, since you can't just do the obvious thing of squishing $Z\setminus X*Y_i$ down to $X$ (you need to also squish down $X*Y_i$ at the same time to be sure this is continuous). $\endgroup$ Oct 20, 2016 at 22:03
  • 1
    $\begingroup$ @EricWofsey Ok, there is a homotopy $g_t:X*Y\to X*Y$ which collapses the portion associated with $[0,1/2]$ onto the $X$ side while stretching the $[1/2,1]$ portion to $[0,1]$ (say $g_t(x,y,s)=(x,y,\max(0,(1+t)s-t))$). Composing any path with $g_t$ induces a homotopy which lets us assume any piece of a loop in $X*Y$ is contained in a $X*Y_i$. $\endgroup$ Oct 21, 2016 at 4:24
  • $\begingroup$ Right, that's basically what I had in mind. $\endgroup$ Oct 21, 2016 at 5:02

Here is a proof using van Kampen's theorem for the fundamental groupoid (see http://www.math.harvard.edu/~hirolee/pdfs/231a-35-van-kampen.pdf, for instance). To start, it is easy to see that $X*Y$ is path-connected, since $X$ is path-connected, and every point of $X*Y$ by definition is on a path to a point of $X$ (thinking of $X*Y$ as $X\coprod Y$ together with a path from each point of $X$ to each point of $Y$).

Now let $U\subset X*Y$ be the image of $X\times Y\times[0,1)$ under the quotient map and $V\subset X*Y$ be the image of $X\times Y\times (0,1]$ under the quotient map. By van Kampen, we can compute the fundamental groupoid $\pi_{\leq 1}(X*Y)$ as the pushout in groupoids of $\pi_{\leq 1}(U)\leftarrow\pi_{\leq 1}(U\cap V)\to\pi_{\leq 1}(V)$.

Let's now take a closer look at $U$, $V$, and $U\cap V$. Note that $U$ deformation-retracts to $X$, by sending $(x,y,s)$ to $(x,y,st)$ as $t$ goes from $0$ to $1$ (this respects the equivalence relation that defines $X*Y$ since $U$ contains no points with $s=1$). Similarly, $V$ deformation-retracts to $Y$, by linearly moving the $s$ coordinate to $1$. Finally, $U\cap V$ is just $X\times Y\times (0,1)$ which is homotopy-equivalent to $X\times Y$.

In particular, it follows any path $\gamma$ in $V$ between points $(x,y,s)$ and $(x',y',s')$ in $U\cap V$ is homotopic to a path in $U\cap V$. Namely, on the $X$-coordinate choose any path from $x$ to $x'$ in $X$ (here we use path-connectedness of $X$), on the $Y$-coordinate take the path given by $\gamma$, and on the $I$-coordinate take the linear path from $s$ to $s'$. This path has the same projection onto $Y$ as $\gamma$, and hence is homotopic to $\gamma$ since the projection to $Y$ is our homotopy equivalence from $V$ to $Y$.

Since $\pi_{\leq 1}(X*Y)$ is the pushout of $\pi_{\leq 1}(V)$ and $\pi_{\leq 1}(U)$ over $\pi_{\leq 1}(U\cap V)$, we conclude that any path in $X*Y$ between two points of $U$ is homotopic to a path contained in $U$ (since all the parts that are in $V$ can be homotoped to be in $U\cap V$). In particular, for any basepoint $p\in U$, the inclusion map $\pi_1(U,p)\to \pi_1(X*Y,p)$ is surjective. But the inclusion $X\to U$ is a homotopy equivalence, and the inclusion $X\to X*Y$ is nullhomotopic since any point of $Y$ gives a cone over $X$ in $X*Y$. It follows that the inclusion $U\to X*Y$ is nullhomotopic, and in particular induces a trivial map on $\pi_1$. Since the inclusion map $\pi_1(U,p)\to \pi_1(X*Y,p)$ is both surjective and trivial, $\pi_1(X*Y,p)$ is trivial. Thus $X*Y$ is simply connected.


Let $(x,y,t),(x',y',t')$ be two points in $X*Y$, and $\alpha:[0,1]\to X$ a path connecting $x$ and $x'$. Then define $$\gamma(s)=\begin{cases} (x,y,(1-3s)t) &\text{if } 0\leq s\leq 1/3\\ (x,y',(3s-1)t') &\text{if } 1/3\leq s\leq 2/3\\ (\alpha(3s-2),y',t') &\text{if } 2/3\leq s\leq 1\\ \end{cases}$$ to get a path joining $(x,y,t)$ and $(x',y',t')$. Thus $X*Y$ is path-connected.

To see that $X*Y$ is simply-connected, let $\gamma:[0,1]\to X*Y$ be a path with $\gamma(0)=\gamma(1)$. Write $\gamma(s)=(x(s),y(s),t(s))$. We want to show that $\gamma$ is null-homotopic. WLOG we can assume $\gamma(0)=\gamma(1)=(x,y,0)$. Let $f(s,r)=(x(s),y(s),rt(s))$ for $r\in [0,1]$. Then $f$ is a homotopy between $\gamma$ and $\gamma'$ defined by $\gamma'(s)=(x(s),y,0)$. We can then define a homotopy $$g(s,r)=\begin{cases} (x,y,2s) &\text{if } s\leq r/2\\ \left(x\left(\frac{s-r/2}{1-r}\right),y,r\right) &\text{if } r/2< s< 1-r/2\\ (x,y,2-2s) &\text{if } s\geq 1-r/2 \end{cases}$$ between $\gamma'$ and $\gamma''$ defined by $$\gamma''(s)=\begin{cases} (x,y,2s)&\text{if } s\leq 1/2\\ (x,y,2-2s)&\text{if } s\geq 1/2 \end{cases}$$ which is clearly null-homotopic. Composing homotopies shows that $\gamma$ is null-homotopic, hence $X*Y$ is simply-connected.

  • $\begingroup$ Ah, there was a typo in the title; I wish to show that the join is simply-connected. But thanks for your construction! $\endgroup$ Aug 27, 2012 at 20:20
  • $\begingroup$ @RoyBen-Abraham I've edit the answer. I believe this shows $X*Y$ is simply connected. $\endgroup$ Aug 27, 2012 at 23:49
  • $\begingroup$ Your $f$ is not well-defined: if $t(s)=1$, then $x(s)$ is not well-defined. $\endgroup$ Oct 20, 2016 at 7:08

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.