enter image description here

I only have difficulties in problem 11.

My efforts:

Follow the hint. We have a sequence of functions {$F_k(t)$}. Each $F_k(t)$ maps $t\in$ [0, 1] to a sequence $(t_1,t_2,...)$ in $\Pi_{n\geq 1}[0,1]_n$. In problem 10, we have constructed a continuous function $f^{(k)}(t)$ from [0, 1] onto $[0, 1]^k$. Now we make reference to a result from a previously solved problem. enter image description here

Let $X=[0,1]^k$ and $Y=\Pi_{n>k}[0,1]_n$. Let $y_0$ be the all-zero sequence in $Y$. Then the map $f:X\rightarrow X\times Y$ defined by $f(x)=x\times y_0$ is continuous. Then $F_k(t)$ is the composite of two continuous functions: $F_k(t)=f(f^{(k)}(t))$. This holds for each fixed $k$. Thus each $F_k(t)$ is continuous. Of course it is not onto due to its 0-tail.

Recall that $[0, 1]^k$ is using the square metric on $\mathbb{R}^k$: $d_k(x,y)=$ max{$|x_1-y_1|,|x_2-y_2|,...,|x_k-y_k|$}. Also recall that $\mathcal{C}(I,I^k)$ is using the corresponding sup metric: $\rho_k(f,g)=$ sup{$d_k(f(t),g(t))|t\in I$}.

Now $\Pi_{n\geq 1}[0,1]_n$ is using product topology. We cannot treat it as a metric space. Accordingly there is no metric for $F_k(t)$. We cannot write something like $\rho(F_7,F_9)$. We cannot prove the convergence of {$F_k(t)$} by saying it is Cauchy. We can only use the open-neighborhood definition of convergence for {$F_k(t)$}. What's the open neighborhood of $F_k(t)$? We can view $F_k(t)$ as a point in the product space $[0,1]\times\Pi_{n\geq 1}[0,1]_n=\Pi_{n\geq 0}[0,1]_n$. Then the convergence of {$F_k(t)$} to $F(t)$ is defined as: Any open neighborhood of $F(t)$ intersects {$F_k(t)$}. More exactly, it is defined as: Given any $n$ and any $\epsilon>0$, {($x,y$)|max{$|x-t|,d_n(y_{1:n},F(t)_{1:n})$}<$\epsilon$} intersects {$F_k(t)$}. That is, given any $n$ and any $\epsilon>0$, there exists $k$ such that $|t-s|<\epsilon$ and $d_n(F_k(s),F(t))<\epsilon$. The convergence cannot be guaranteed by the construction of $f^{(k)}(t)$. We can find a $F_k(s)$ having the same 1-to-$n$ coordinates as F(t), but $|t-s|$ may be greater than $\epsilon$.

------------------------Updated according to the hints given by Brian M. Scott------------------------

Let $h^{(k)}(t)$ be the $k$-time composition of $h(t)$, e.g., $h^{(3)}(t)=h(h(h(t)))$.





Given $n$, $t_n^{(k)}=0$ for $k<n$. We claim that $t_n^{(k)}=g(h^{(n-1)}(t))$ for $k>n$. We show this by induction on $n$.

By construction, $f^{(k+1)}(t)=(g(t),f^{(k)}(h(t)))=(g(t),g(h(t)),f^{(k-1)}(h^2(t)))=(g(t),g(h(t)),g(h^{(2)}(t)),f^{(k-2)}(h^3(t)))=...$

The $n=1$ case is that $t_1^{(k)}=g(t)$. This is obvious from the construction.

Assume it is true for all $n$. By construction, $t^{(k)}_{n+1}=h(t)^{(k)}_n=g(h^{(n-1)}(h(t)))=g(h^{(n)}(t))$.

We have shown that $t_n^{(k)}$ is constant after $k>n$.

Next we show $F$ is continuous. We only need to show: Given any open neighborhood $U$ of $\langle t_n:n\geq 1\rangle$, $F^{-1}(U)$ is open.

Since $\prod_{n\geq 1}[0,1]_n$ is using product topology, we can take $U$ to be $$U=\prod^N_{n=1}\big((t_n-\epsilon_n,t_n+\epsilon_n)\cap [0,1]_n\big)\times\prod_{n\geq N+1}[0,1]_n$$ for any fixed $N$. Since $f^{(N)}$ is continuous when $[0,1]^N$ is using square metric topology and the ball $$B(\langle t_n,n=1,...,N\rangle,\min\{\epsilon_1,\ldots,\epsilon_N\}, d_N)\subset\prod^N_{n=1}\big((t_n-\epsilon_n,t_n+\epsilon_n)\cap [0,1]_n\big)\;,$$ $f^{(N)}$ is also continuous when $[0,1]^N$ is using box topology. Thus $F^{-1}(U)=(f^{(N)})^{-1}\left(\prod^N_{n=1}\big((t_n-\epsilon_n,t_n+\epsilon_n)\cap [0,1]_n\big)\right)$ is open.

Finally we need to show $F$ is surjective: Given any $\langle t_n,n\geq 1\rangle\in\prod_{n\geq 1}[0,1]_n$, there exists a $t\in[0,1]$ such that $F(t)=\langle t_n,n\geq 1\rangle$. For any fixed $N$, we can find a $t$ such that $g(h^{(n-1)}(t))=t_n$ for all $n<N$ by construction of $F(t)$. Since $t$ and $t_n$ remain constant as we increase $N$, we have found the desired $t$.


The product topology on $\prod_{n\ge 1}[0,1]_n$ actually is metrizable: this is true of any product of countably many metrizable spaces. However, you don’t need that here.

HINT: For $t\in[0,1]$ and $k\ge 1$ let $F_k(t)=\left\langle t_n^{(k)}:n\ge 1\right\rangle$.

  • Show that for each $n\ge 1$ the sequence $\left\langle t_n^{(k)}:k\ge 1\right\rangle$ is eventually constant and therefore converges to some $t_n\in[0,1]$. (In fact you should find that $t_1=g(t)$, $t_2=g(h(t))$, $t_3=g(h(h(t)))$, and so on.)

Define $F:[0,1]\to\prod_{n\ge 1}[0,1]_n:t\mapsto\langle t_n:n\ge 1\rangle$.

  • Show that $F$ is continuous.
  • Show that $F$ is a surjection.
  • $\begingroup$ Please see the last paragraph of my updated question description. Is my proof of surjection correct? $\endgroup$ – Junk Warrior Jul 11 '20 at 22:34
  • $\begingroup$ @JunkWarrior: Yes, that looks okay. I fixed up some of the MathJax to make it easier to read. To get real angle brackets you can use \langle ($\langle$) and \rangle ($\rangle$); the max and min operators are \max ($\max$) and \min ($\min$); and the product symbol is \prod ($\prod$), just as the summation symbol is \sum ($\sum$). $\endgroup$ – Brian M. Scott Jul 11 '20 at 23:35

Another route to the same fact: we consider the Cantor set $\mathbf{C} \subseteq [0,1]$. This is well known to be homeomorphic to $C=\{0,1\}^{\Bbb N}$, the countable Cantor cube. The well-known Cantor function maps $C$ onto $[0,1]$ (sending a sequence $(x_n)_n \in C$ to $\sum_n \frac{x_n}{2^{n+1}}$). It's clear that $C^{\Bbb N} \simeq C$ (a reshuffle of the countably many coordinates on both sides) and so combining the mentioned homeomorphisms with the countable power of the Cantor function, we get a map $\mathbf{C} \to [0,1]^{\Bbb N}$ and we extend this by Tietze'e extension theorem to a surjective continuous map $[0,1] \to [0,1]^{\Bbb N}$.

Bonus fact: any compact connected and locally connected metric space is the continuous image of $[0,1]$ (one of the main theorems on so-called Peano continua). And $[0,1]^{\Bbb N}$ is metrisable as any countable product of metric spaces is (e.g. see here), and standard theorems tell us it's compact, connected and locally connected.


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.