metric on infinite product, continuous

192 Views Asked by At

Let $(X_k,d_k)_{k\in\mathbb{N}}$ be a family of metric spaces. Let $X=\prod_{k=1}^\infty X_k$ be equipped with the metric $$d(u,v)=\sum_{k=1}^\infty 2^{-k}\overline{d_k}(u_k,v_k),$$ where $\overline{d_k}=\min\{1,d_k\}$

Show that a function $f:Y\to X$ from a metric space $(Y,d_Y)$ is continuous if and only if for every $k\in\mathbb{N}$ the composition $f_k=\operatorname{pr}_k\circ f: Y\to X_k$ is continuous, where $\operatorname{pr}_k\colon X\to X_k$ is the canonical projection.

During the lecture, a similar statement was proven for finite products.

However, the proof was not correct, and I tried to prove the statement above similarly to the answer of Theo Bendit to the question universal property of product metric, proof explanation.

But I don't think this works here. My argument is as follows:

I have already proven that $\operatorname{pr}_k$ (the projection on the $k$-th coordinate) is continuous, so one implication is trivial.

For the other implication, I argued as follows: Suppose that $f_k$ is continuous for every $k\in\mathbb{N}$. Let $W\subseteq X$ be open. We want to show that $f^{-1}(W)$ is open.

Since $W\subseteq X$ is open, we have for every $w:=(w_1,w_2,\ldots)\in W$ an $\varepsilon > 0$ such that $B_\varepsilon^d(w)\subseteq W$.

Now let $V:=\prod_{k\in\mathbb{N}} B_\varepsilon^{d_k}(w_k)$. We want to show that $V\subseteq B_\varepsilon^d(w)$.

So let $v\in V$. Let us show that $v\in B_\varepsilon^d(w)$.

We have $$d(v,w)=\sum_{k=1}^\infty 2^{-k}\overline{d_k}(v_k,w_k)\stackrel{!}{\leq} \sum_{k=1}^\infty 2^{-k}d_k(v_k,w_k)<\sum_{k=1}^\infty 2^{-k}\cdot \varepsilon=\varepsilon.$$

Hence $v\in B_\varepsilon^d(w)$.

Now the problem: I would like to show that $$\bigcap_{k\in\mathbb{N}} f_k^{-1}(B_\varepsilon^{d_k}(w_k))\subseteq f^{-1}(W),$$ but the intersection of infinitely many open sets is not necessarly open. So this should not work.

How can this be fixed?

Thanks in advance.

1

There are 1 best solutions below

0
On BEST ANSWER

The main thing you're missing is that the "tail" of the series defining $d$ goes to zero: all $\overline{d_k}$ are $\leq 1$, so for all $N$, $$\sum_{k=N+1}^\infty 2^{-k}\overline{d_k}(u_k,v_k)\leq\sum_{k=N+1}^\infty 2^{-k}=2^{-N}$$ so given $\epsilon>0$, you just need to take $N$ very large (depending on $\epsilon$) so that $2^{-N}<\epsilon/2$. If $y'$ is very close to $y$, then $f_k(y')$ will be very close to $f_k(y)$.

In fact, we can even make $d_k(f_k(y'),f_k(y))<\epsilon/(2N)$, as long as $d(y',y)<\delta_k$ for some $\delta_k$ depending only on $y$, on $k$, on $N$ and on $\epsilon$. So as long as $d(y',y)<\delta:=\min\left\{\delta_1,\ldots,\delta_N\right\}$ (this $\delta$ depends only on $\epsilon$ and $N$), we will have $\sum_{k=1}^N2^{-k}\overline{d_k}(f_k(y'),f_k(y))<\epsilon/2$.

Putting the two inequalities above together, we conclude $d(f(y'),f(y))<\epsilon$, whenever $d(y',y)<\delta$, and $\delta$ deends only on $y$ and on $\epsilon$ (and some choice of $N$ which depends only on $\epsilon$, so it does not matter).


An easier solution uses sequential continuity:

Let $(x_n)_{n=1}^\infty$ be a sequence in $X$. Each element $x_n$ is itself a sequence, so let us write its terms with superscripts just to avoid confusion: $x_n=(x_n^k)_{k=1}^\infty$. Consider another element $x=(x^k)_{k=1}^\infty$ of $X$.

Exercise: Prove that $x_n$ converges to $x$ in $X$ if and only if for every $k$, $x_n^k$ converges to $x^k$ in $X_k$.

Hint: For any two elements $a=(a^k)_k$ and $b=(b^k)_k$ of $X$ and any $N$, we have $$d(a,b)\leq \sum_{k=1}^N d(a^k,b^k)+\sum_{k=N+1}^\infty 2^{-k}=\sum_{k=1}^N d(a^k,b^k)+2^{-N}$$


To answer your question: If all $f_k$ are continuous, suppose $y_n\to y$ in $Y$. Then for each $k$, $f_k(y_n)\to f_k(y)$ in $X_k$, so $f(y_n)\to f(y)$ in $X$ by the exercise above.