Let $\mathbb{R}^{\infty}$ be the vector space of infinite sequences $(\alpha_1,\alpha_2,\alpha_3,\dots)$ of real numbers as a natural generalization of $\mathbb{R}^k$, $k\in\mathbb{N}^*$. Let $\boldsymbol{C}$ be the subset of $\mathbb{R}^{\infty}$ defined by, $$\boldsymbol{C} := \left\{\boldsymbol{p} = (p_1, p_2, p_3, \dots) \in \mathbb{R}^{\infty}\big/ \forall ~k, p_k \in [0, 1] \text{ and } \sum_{k = 1}^{\infty} p_{k} \leq 1\right\}.$$ I make these statements. Tell me if I am wrong.

  1. $\boldsymbol{C}$ is a compact and convex nonempty subset of the infinite dimensional space $\mathbb{R}^{\infty}$ for the the infinite norm defined by, $$||\boldsymbol{p}||_{\infty} = \max_k p_{k}.$$
  2. Schauder's fixed point Theorem (generalization of Brouwer's fixed point Theorem to an infinite dimensional space) can be applied: any continuous mapping from $\boldsymbol{C}$ to itself has a fixed point.
  • $\begingroup$ It's not compact. Set $p^{(n)} = (\delta_{in})_{i \in \mathbb{N}}$. You will not be able to find a convergent subsequence. $\endgroup$ – Peter Morfe Jan 25 at 23:26
  • $\begingroup$ Thank you @PeterMorfe. I don't get it. What is the sequence you set? It seems that you have a sequence of sequence. $\endgroup$ – Ari.stat Jan 25 at 23:30
  • $\begingroup$ If $\sum_{i = 1}^{\infty}\delta_{in} = 1$ for all $n$ then the this is also true for the limit $(\bar{\delta}_i)$: $\sum_{i = 1}^{\infty}\bar{\delta}_{i} = 1$. $\endgroup$ – Ari.stat Jan 25 at 23:36

Define a sequence $(p^{(n)})_{n \in \mathbb{N}}$ in $C$ by $p^{(n)} = (\delta_{in})_{i \in \mathbb{N}}$. In other words, \begin{equation*} p^{(n)}_{i} = \left\{ \begin{array}{r l} 1, & \text{if} \, \, i = n, \\ 0, & \text{otherwise.} \end{array} \right. \end{equation*}

$(p^{(n)})_{n \in \mathbb{N}}$ does not have any accumulation points in $C$ with the given norm topology. The idea is this: $\|p^{(n)} - p^{(k)}\|_{\infty} = 1$ for all $n,k \in \mathbb{N}$ so no subsequence is Cauchy.

Edit: Given $m \in \mathbb{N}$, if we define $C^{m}$ by \begin{equation*} C^{m}_{*} = \{x \in \mathbb{R}^{\infty} \, \mid \, \sum_{k = 1}^{\infty} |x_{k}| \leq 1, \, \, |x_{k}| \leq \frac{m}{k} \}, \end{equation*} then $C^{m}_{*}$ is compact in $\mathbb{R}^{\infty}$ with the norm $\|\cdot\|_{\infty}$ given above. Note that the set $C^{m}$ defined by Ari.stat below is simply $C^{m} = C^{m}_{*} \cap [0,1]^{\infty}$. Since $[0,1]^{\infty}$ is a closed subset of $\mathbb{R}^{\infty}$ with respect to $\|\cdot\|_{\infty}$, $C^{m}$ is a closed subset of $C^{m}_{*}$ and, thus, is itself compact.

To see this, first, notice that if $x \in C^{m}_{*}$, then $\|x\|_{\infty} \leq \sum_{k = 1}^{\infty} |x_{k}| \leq 1$.

To see that $C^{m}_{*}$ is compact, I will prove that any sequence in $C^{m}$ has a subsequence that converges. Suppose that $(x^{(n)})_{n \in \mathbb{N}} \subseteq C^{m}_{*}$. Since $|x^{(n)}_{k}| \leq 1$ independently of $n$ and $k$, we can employ a diagonalization argument to find a subsequence $(n_{j})_{j \in \mathbb{N}} \subseteq \mathbb{N}$ and a sequence $x \in \mathbb{R}^{\infty}$ so that, for each $k \in \mathbb{N}$, $\lim_{j \to \infty} x_{k}^{(n_{j})} = x_{k}$. In fact, we can show that $\lim_{j \to \infty} \|x^{(n_{j})} - x\|_{\infty} = 0$ must hold.

Indeed, given $\epsilon > 0$, if we choose $K_{0} \in \mathbb{N}$ such that $\frac{m}{k} \leq \epsilon/2$ for each $k \geq K_{0}$, then the inequality $\max\{|x^{(n)}_{k}|, |x_{k}|\} \leq \frac{m}{k}$ implies \begin{equation*} \|x^{(n_{j})} - x\|_{\infty} \leq \max \{ |x^{(n_{j})}_{1} - x_{1}|, \dots, |x^{(n_{j})}_{K_{0}} - x_{K_{0}}|, \epsilon \}. \end{equation*} Therefore, by pointwise convergence, $\limsup_{j \to \infty} \|x^{(n_{j})} - x\|_{\infty} \leq \epsilon$. We conclude by sending $\epsilon \to 0^{+}$.

  • $\begingroup$ Thank you. So what is the contradiction here with the compactness? I understand that $(p^{(n)})_{n\in\mathbb{N}}$ does not have a limit in $C$. But anyway, $(p^{(n)})_{n\in\mathbb{N}}$ does not have a limit in $\mathbb{R}^{\infty}$. $\endgroup$ – Ari.stat Jan 26 at 0:22
  • 1
    $\begingroup$ If $C$ were compact, then any infinite subset would have to have at least one accumulation point. (This is true whenever we are working with compact sets in metric spaces.) $\endgroup$ – Peter Morfe Jan 26 at 0:46
  • $\begingroup$ Woww I got your point. Very useful. What if I defined the space as $$\boldsymbol{C^m} := \left\{\boldsymbol{p} = (p_1, p_2, p_3, \dots) \in \mathbb{R}^{\infty}\big/ \forall ~k, p_k \in [0, 1]; ~ p_k \leq \frac{m}{k}; \text{ and } \sum_{k = 1}^{\infty} p_{k} \leq 1\right\}?$$ I do not want to change my post because your answer is useful. Can you prove in another answer if the statements are true for some $m > 0$ with this new space? $\endgroup$ – Ari.stat Jan 26 at 13:52
  • 1
    $\begingroup$ @Ari.stat, I edited the answer to address the question in your last comment. $\endgroup$ – Peter Morfe Jan 30 at 19:36

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.