I am having trouble understanding the product of CW complexes. I know how to actually do the computations and all, I just don't understand how exactly it works.

So here's my questions specifically: If $X,Y$ are CW-complexes then say $e,f$ are $p,q$ cells on $X,Y$ respectively, then we know that $e \times f$ will be a $p+q$ cell in $X \times Y$. But this cell we have to think of as $D^{p+q}$ with some identification on the boundary sphere. But what we have here is $(D^p \times D^q)$. So I guess we need to know that we have homeomorphism of pairs $(D^{p+q},S^{p+q-1}) \cong (D^p \times D^q, S^{p-1} \times D^q \cup D^p \times S^{q-1})$. But that's what I do not get. How exactly do this homeomorphism work?

I kinda see it in the case $p=q=1$ (it's just that a square and a disc are homeomorphic with canonical identifications in the boundaries). But I am having trouble defining it or actually visualizing it in higher dimensions. Any help?


  • $\begingroup$ I'm not sure I can help you visualize it in general, but we can at least manage with $p = 2$ and $q = 1$. Then $D^2 \times D^1$ is a solid cylinder, and the boundary is $S^1 \times D^1$, a vertical tube, together with $D^2 \times S^0$, the two end caps. The boundary operator on manifolds with boundary is a derivation in this fashion in general: $\partial(M \times N) = (\partial M \times N) \cup (M \times \partial N)$. $\endgroup$ – jdc May 13 '14 at 23:08
  • 1
    $\begingroup$ Oh OK. So you're saying that just because $D^{p+q} \cong D^p \times D^q$ the iso on the boundaries will follow? Huh that was really simple after all! Thanks! $\endgroup$ – Bernard May 13 '14 at 23:26
  • $\begingroup$ If you think of it in terms of manifolds with boundary/corners, the points in the product that don't have open ball neighborhoods have to have at least one coordinate on the boundary of one of the factor manifolds. $\endgroup$ – jdc May 13 '14 at 23:29

You can think of $D^n$ as the homeomorphic cube $I^n$. This way, the product $$\left(D^k\times D^l,\ \partial D^k× D^l\cup D^k×∂D^l\right)\\ \cong\left(I^k×I^l,\ ∂I^k×I^l\cup I^k×∂I^l\right)\\ =\left(I^{k+l},∂\left(I^k×I^l\right)\right)\\ \cong \left(D^{k+l},∂\left(D^k×D^l\right)\right)$$

The homeomorphism between $D^k\cong I^k$ i given by $$x\mapsto \dfrac{x\cdot||x||_2}{||x||_\infty}$$

The characteristic map is $\Phi_{\alpha,\beta}=Φ_{α}×Φ_β:D^k×D^l\to X×Y$

  • Note that since $X,Y$ are Hausdorff, so is $X\times Y$.
  • The images of the interiors $Φ_{α,β}\left(\text{int}D^{k+l}\cong\text{int}D^k×\text{int}D^l\right)$ partition $X\times Y$.
  • For each product cell $e_α×e_β$, the image of $∂\left(D^k×D^l\right)= ∂I^k×I^l\cup I^k×∂I^l$ is in finitely many cells of dimension less than $k+l$
  • If the product topology on $X×Y$ is such that each set $A$ is closed if $A\cap \overline{e_{α,β}}$ is closed in $\overline{e_{α,β}}$ for each cell, then all conditions in the implicit definition are satisfied, so $X×Y$ will be a CW complex.
  • $\begingroup$ The last condition generally not holds, but it holds for certain case like when either $X$ or $Y$ is locally compact. But I got stuck when proving it. Any hints? $\endgroup$ – Chen M Ling May 26 '16 at 18:52
  • 1
    $\begingroup$ @ChenMLing The idea is that when $X\times Y$ is a $k$-space (also called compactly generated space), then the product topology is the topology as a CW complex, see math.stackexchange.com/questions/248639/…. This holds when $X$ or $Y$ is locally compact since the product of a $k$-space with a locally compact space is again a $k$-space. $\endgroup$ – Stefan Hamcke May 27 '16 at 0:35
  • $\begingroup$ Yes. There is such theorem in Hatcher that the product of a locally compact space and a k-space is again a k-space. Since $X$ is a CW complex, then it is directly a k-space. But now the result gives $X\times Y$ is a k-space, this is where I got stuck: how to show that it is CW? $\endgroup$ – Chen M Ling May 27 '16 at 10:43
  • $\begingroup$ See my answer for math.stackexchange.com/questions/248639/…. It shows that when $X\times Y$ is a $k$-space, then it is a CW complex. @ChenMLing $\endgroup$ – Stefan Hamcke May 27 '16 at 17:35
  • $\begingroup$ Oh I missed that. Thanks @StefanHamcke $\endgroup$ – Chen M Ling May 28 '16 at 6:09

I hope the following picture of a solid torus with the cells labelled will help:


It is taken from Nonabelian Algebraic Topology (pdf available there), which also compares algebraic modelling of products of cells as in your question and in terms of cubes; these are much easier to understand since the $m$-cube $I^m$ is the topological product of $m$ copies of the unit interval $I=[0,1]$ and so $I^{m+n} \cong I^m \times I^n$. The relation between cells and cubes is used a lot in algebraic topology. The natural cell structure of $I^m$ is more complicated than that of the $m$-cell $D^m= e^0 \cup e^{m-1} \cup e^m$, but you gain in other ways, so one needs both models.


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.