5
$\begingroup$

Suppose I consider $C_0(\mathbb{N})$ consisting of function on the natural numbers vanishing at $\infty$. For an irrational $1<\alpha<2$, let $p_{m\alpha}(\cdot)$ be the function $p_{m\alpha}(n)= 1/n^{m\alpha}$. By the Stone-Weierstrass theorem, linear combinations of $\{p_{m\alpha}\},m\geq 1$ is dense in $C_0(\mathbb{N})$. If I only retain the fractional part $\{m\alpha\}$ of each $m\alpha$, is the span of $\{p_{\{m\alpha\}}\},m\geq 1$ still dense in $C_0(\mathbb{N})$?

$\endgroup$
9
  • $\begingroup$ When you write $\{p_{m\alpha}\},m\geq 1$, you mean $\{p_{m\alpha}\mid m\geq 1\}, right?$ $\endgroup$
    – Goldstern
    Commented May 5, 2023 at 8:47
  • $\begingroup$ Yes, the sets of functions should be $\{p_{m\alpha},m\geq 1\}$ and $\{p_{\{m\alpha\}},m\geq 1\}$. $\endgroup$
    – F J
    Commented May 5, 2023 at 9:04
  • $\begingroup$ (Add to comment: even $p^\alpha$ alone should separate the points and generate your algebra, essentially, modulo unit) $\endgroup$
    – Ewrt Wert
    Commented May 5, 2023 at 11:23
  • $\begingroup$ (Add to comment: I read too fast, but as your functions separate the points, the only question is if norm closure of span of $p^{\{mα\}}$ forms algebra -> this might be questioned) (How do you write $e^{-n}$ or only $n^{-2}$ approximately as linear span of your functions $1/n^{u}$ with $0 < u < 1$?) $\endgroup$
    – Ewrt Wert
    Commented May 5, 2023 at 12:27
  • $\begingroup$ But the exponent is constant for each function $\endgroup$
    – Ewrt Wert
    Commented May 5, 2023 at 12:38

2 Answers 2

8
$\begingroup$

We can use the Müntz–Szász theorem, which states that if $\Lambda = \{0 = \lambda_0 < \lambda_1 < \cdots\}$ then the span of $\{x^\lambda : \lambda \in \Lambda\}$ is dense in $C[0,1]$ if and only if $\sum_{k=1}^\infty 1/\lambda_k = \infty$. Note: Wikipedia restricts $\Lambda \subset \mathbb N$, but this is not needed. For example see [Rudin, Real and Complex Analysis, Theorem 15.26] or [Almira, Muntz type Theorems I, Theorem 1].

Here we take $\Lambda$ to be the union of $\{0\}$ and an infinite subset of $\{\{m\alpha\} : m \in \mathbb N\}$ accumulating at $1$. The theorem then tells us that $\{x^\lambda : \lambda \in \Lambda\}$ is dense in $C[0,1]$.

Let $f \in C_0(\mathbb N)$. Define $F \in C[0,1]$ by $F(0) = 0$, $F(1/n) = f(n)$ for $n \in \mathbb N$, and linearly interpolating. Then we can uniformly approximate $F(x)$ by some linear combination $\sum_{i=0}^K a_i x^{\lambda_i}$. By evaluating at $0$ we find that $a_0$ must be small, so we can throw it away. Now by restricting to the values $F(1/n)$, it follows that $f(n)$ is uniformly approximated by $\sum_{i=1}^K a_i / n^{\lambda_i}$.

$\endgroup$
2
  • 1
    $\begingroup$ I had the same idea of applying MS thm, but I forgot the obvious fact that here $\{m\alpha\}$ are dense in [0,1]! $\endgroup$ Commented May 5, 2023 at 14:18
  • $\begingroup$ Thank you for the answer. $\endgroup$
    – F J
    Commented May 6, 2023 at 0:33
4
$\begingroup$

Let's change variable and use $x=1/n$, so the question reads: when $\{1, x^{\lambda_1},x^{\lambda_2},\dots\}$ has dense linear span in the space of the continuous functions on $K$ vanishing at $0$, for $K:=\{0,1/n:n\in\mathbb N\}$ and $\lambda_m:=\{m\alpha\}$.

[useless assumption] At least for $\alpha$ a quadratic algebraic irrational, we have affirmative answer. In this case, by diophantine approximation, there is $C>0$ such that $\lvert\alpha-p/q\rvert>C/q^2$ holds for all $p\in\mathbb Z$ and $q\in\mathbb Z_+$, so that $\{q\alpha\}>C/q$ for all $q\ge1$.

Recall the full Müntz–Szász theorem: for a sequence of positive numbers $\{\lambda_q\}_{q\ge1}$ the set $\{1, x^{\lambda_1},x^{\lambda_2},\dots\}$ has dense span in $C^0[0,1]$ if ad only if $\displaystyle \sum_{q\ge1}\frac{\lambda_q}{1+\lambda_q^2}=+\infty$ (a condition that reduces to simply $ \sum_{q\ge1} {\lambda_q} =+\infty$ for bounded positive sequences, which is the case of $\lambda_q:=\{\alpha q\}$). Here, with change of variable $x=1/n$, we have in particular that $\{ n^{-\{ q\alpha\}}: q\ge1\}$ is dense in the space $c_0$.

[edit]: In fact the diophantine thing here is totally unnecessary, for of course here $\lambda_q$ are dense in $[0,1]$, which already ensures the assumption of the MS theorem, as in Sean Eberhard's answer!

$\endgroup$
1
  • $\begingroup$ Thank you for the answer. $\endgroup$
    – F J
    Commented May 6, 2023 at 0:33

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .