5
$\begingroup$

It is well-known that the harmonic series is not summable. In some sense this means that it takes a lot of rather large values.

We define the operator $F_{\varepsilon}: \ell^{\infty}(\mathbb N) \rightarrow [0,\infty]$ by $$F_{\varepsilon}(x) = \sum_{i=1}^{\infty} 2^{-\varepsilon \vert x_i \vert^{-1}} \text{ for }\varepsilon>0.$$

Now consider a positive summable sequence $x$ and the harmonic sequence $(1/n)_n$. Intuitively, the slow decay of the harmonic series should imply that it converges slower than anything summable (for most of it).

Therefore, I ask: Is it true that for any positive summable sequence $x$

$$\limsup_{\varepsilon \downarrow 0} \frac{F_{\varepsilon}(x)}{F_{\varepsilon}((1/n))} \le 1?$$

Please just let me know if you have any questions.

$\endgroup$
9
  • 2
    $\begingroup$ What about $x=(2/n)_n$? $\endgroup$ Commented Mar 18, 2021 at 5:40
  • $\begingroup$ @EthanDlugie that sequence is not summable though? $\endgroup$
    – Sascha
    Commented Mar 18, 2021 at 11:25
  • $\begingroup$ You are asking in the wrong forum. $\endgroup$ Commented Mar 18, 2021 at 12:02
  • 1
    $\begingroup$ @AlexandreEremenko: Since, for example, $2^{-\epsilon/t} \le C(\epsilon) t^2$ for $t > 0$, $F_\epsilon(x)$ is finite for every square-summable sequence. :-) $\endgroup$ Commented Mar 18, 2021 at 14:04
  • 2
    $\begingroup$ @GeraldEdgar: While this is indeed a relatively simple problem, I have seen many more basic questions answered at this forum. $\endgroup$ Commented Mar 18, 2021 at 14:06

1 Answer 1

6
$\begingroup$

$\newcommand\ep\varepsilon\newcommand\de\delta$ Let us show more: \begin{equation*} \frac{F_{\ep}(x)}{F_{\ep}((1/n))}\to0\tag{$*$} \end{equation*} (as $\ep\downarrow0$). Indeed, \begin{equation*} F_{\ep}((1/n))=\sum_{i=1}^\infty 2^{-\ep i}=\frac{2^{-\ep}}{1-2^{-\ep}}\asymp\frac1\ep. \tag{0} \end{equation*}

On the other hand, take any positive $x_n$'s such that $\sum_1^\infty x_n<\infty$. For each natural $k$, let $J_k$ denote the set of all natural $n$ such that $\frac1k\le x_n<\frac1{k-1}$, where $\frac1{k-1}:=\infty$ for $k=1$: \begin{equation*} n\in J_k\iff\frac1k\le x_n<\frac1{k-1}. \end{equation*} Then the $J_k$'s partition the set of all natural numbers. Moreover, the condition $\sum_1^\infty x_n<\infty$ implies \begin{equation*} \sum_{k=1}^\infty|J_k|/k<\infty, \tag{1} \end{equation*} where $|J_k|$ is the cardinality of $J_k$. In particular, it follows that $|J_k|<\infty$ for all $k$. Further, $2^{-\ep/x_n}<2^\ep\times2^{-\ep k}$ for $n\in J_k$. So, \begin{equation*} F_{\ep}(x)<2^\ep\sum_{k=1}^\infty 2^{-\ep k}|J_k|. \end{equation*} Take now any real $\de>0$ and, in view of (1), let $k_\de$ be a natural number such that \begin{equation*} \sum_{k\ge k_\de}|J_k|/k<\de. \end{equation*} Let $c_\de:=\sum_{k=1}^{k_\de-1}|J_k|$. Then \begin{align*} F_{\ep}(x)&<c_\de+2^\ep\sum_{k\ge k_\de} k2^{-\ep k}|J_k|/k \\ & \le c_\de+2^\ep\max_{k\ge1}(k2^{-\ep k})\,\sum_{k\ge k_\de} |J_k|/k \\ & <c_\de+2^\ep\frac1{\ep e\ln2}\,\de \\ & <c_\de+\de/\ep \end{align*} if $\ep\in(0,1/2)$. So, \begin{equation} \limsup_{\ep\downarrow0}\frac{F_{\ep}(x)}{1/\ep}\le\de, \end{equation} for every real $\de>0$. Now ($*$) follows, in view of (0).

$\endgroup$
3
  • 2
    $\begingroup$ Incidentally: if it were the case that the $\limsup$ of the ratio is bounded by 1 for ALL absolutely summable sequences $(x_n)$, then necessary the limit is 0 for all such sequences. This is because given a sequence $(x_n)$, if you form $(y_n)$ by listing each term of $(x_n)$ twice, then $y$ is still absolutely summable and $F_\epsilon(y) = 2 F_\epsilon(x)$. Hence it is in fact necessary to prove the stronger statement. $\endgroup$ Commented Mar 18, 2021 at 17:18
  • 1
    $\begingroup$ @WillieWong : Good point! $\endgroup$ Commented Mar 18, 2021 at 18:05
  • $\begingroup$ @IosifPinelis thanks a lot for your answer. I started wondering about the following, which I understand might have to be a separate question, but perhaps there is an easy work-around: Take $x_n=(1/n^2)$ and $z_n=(n). $ Now, define $F_{\varepsilon}(y) = \sum_n 2^{-\varepsilon y_n}.$ Is it true that among all $y=(y_n)$ positive such that $\sum_n x_n y_n$ is finite, we have $\liminf_{\varepsilon \to 0} \frac{F_{\varepsilon}(y)}{F_{\varepsilon}(z)} >0$? $\endgroup$
    – Sascha
    Commented Mar 21, 2021 at 11:27

You must log in to answer this question.

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