1

Give a direct proof of Tychonov's theorem: If $(X_n, d_n)$ is compact, then $\left(\prod_{n\geq1} X_n, d\right)$ is compact.

Johan
  • 4,072

1 Answers1

5

Let $d_n'=d_n/(1+d_n)$. The metrics $d_n$ and $d_n'$ generate the same topology. Define a metric on the product space by $d^*\big((x_n),(y_n)\big)=\sum_n 1/2^n d_n'(x_n,y_n)$. One can chech that $d^*$ indeed metrizes the product topology. So the countable product is metrizable and compactness and sequential compactness coincide.

We show that the product is sequentially compact. Let $(x_n)$ be a sequence of points, i.e. sequences, in the product space. Let $(y_n^1)$ be a subsequence that converges in the first coordinate, $(y_n^2)$ a further subsubsequence that converges in the second coordinate, and, by construction also in the first coordinate. Continuing this way, we get a sequence $\big((y_n^1),(y_n^2),(y_n^3),\ldots\big)$ of subsequences. We construct now a convergent subsequence of all these subsequences as $(y_1^1,y_2^2,y_3^3,\ldots)$. This sequence converges in every coordinate of the product and therefore in the product topology.

  • Very nice! thanks! – Johan Nov 18 '12 at 18:03
  • @Michael Greinecker: I'm sorry to bother you but I'm clearly missing something in your proof above, How can a subsequence converge to two different coordinates. This is what I'm thinking when you say: "and by construction also in the first coordinate". I think I'm not understanding the setup. If you have a sequence ( or a subsequence for that matter ), does it not have to converge to one number ? Thanks. – mark leeds May 11 '22 at 04:31
  • @markleeds Consider the sequence in $\mathbb{R}^2$ whose $n$th term is $(1/n,(-1)^n)$. It converges in the first coordinate (in $\mathbb{R}$) but not in the second. – Michael Greinecker May 11 '22 at 05:34
  • Oh, So each subsequence really has $n$ subsequences where $n$ is the number of dimensions in the respective product space. If that's the case, I get it. Thanks. – mark leeds May 11 '22 at 17:00
  • Hi Michael: I just had one more question and then that will be the end of my bothers. I was wondering ( it's hard for me to tell ) if the argument on page 44 at the link below, starting with "subsequently, taking sub subsequences, is equivalent to the argument that you gave you in this thread ? If so, then it's all coming together for me because I understand your explanation so the equivalence makes things understandable. thank you so much for your help. https://wt.iam.uni-bonn.de/fileadmin/WT/Inhalt/people/Patrik_Ferrari/Lectures/SS16StochProc/wt2-new.pdf – mark leeds May 12 '22 at 03:42
  • Also, ( assuming Michael says yes ), for others coming to this thread in the future, Michael's argument above combined with the full proof at the link immediately above, is the clearest explanation of kolmogorov's existence theorem that I have found and I have looked hard. – mark leeds May 12 '22 at 03:47
  • Shouldn't it be $d' := d/(1+d)$? – Pedro G. Mattos May 02 '24 at 08:17
  • 1
    @PedroG.Mattos Actually, the subscript $n$ should have been there. Fixed it now; thanks for bringing it to my attention. – Michael Greinecker May 02 '24 at 09:34
  • Sure, no problem. But I meant $d/(1+d)$ instead of $1/(1+d)$. – Pedro G. Mattos May 02 '24 at 09:47
  • 1
    @PedroG.Mattos Got it now; thank you! – Michael Greinecker May 02 '24 at 09:57