Let $(x_n)$ be a monotonically decreasing sequence of positive real numbers that is also summable.

Let $(y_n)$ be a sequence of positive real numbers such that $\sum_n x_n y_n$ converges.

Let $(z_n)$ be a monotonically increasing sequence of positive real numbers such that $\sum_n x_n z_n =\infty.$

Assume that the sequences $y_n$ and $z_n$ are such that $2^{-\varepsilon y_n}$ and $2^{-\varepsilon z_n}$ are summable for every $\varepsilon>0.$ Does it follow that there is some $\delta>0$ such that

$$ \sum_n \Big(2^{-\varepsilon y_n}-2^{-\varepsilon z_n}\Big) \ge 0 \text{ for all } \varepsilon \in (0,\delta)?$$

The motivation for this statement to be true is that $z_n$ should be larger most of the time than $y_n$ and we capture this most of the time by taking $\varepsilon$ small.

Please let me know if you have any comments, questions or remarks.


1 Answer 1


No. For example $x_n=2^{-n}$, $y_n=n$, and we now keep $z_n$ constant on long intervals. More precisely, consider first $\epsilon=1$, and set $z_1= \ldots = z_{N_1}=c_1$, with $N_1$ taken so large that $N_1 2^{-1\cdot 1}>\sum 2^{-1\cdot n}$.

We then continue in the same way: let $z_{N_1+1}=\ldots = z_{N_2}= c_2$, and again take $N_2$ so large that the inequality you want fails for $\epsilon=1/2$ etc.

Here we must choose the $c_n$ such that $\sum 2^{-n}z_n=\infty$ and $\sum 2^{-\epsilon z_n}<\infty$. I'll take $c_n=2^{N_{n-1}}$ to guarantee the first property. If the $n$th step deals with $\epsilon=1/n$, then, since $\sum n2^{-\epsilon n} \simeq 1/\epsilon^2$, I can take $N_n=n^3 2^{N_{n-1}/n}$ to make sure your inequality fails.

With these choices in place, the final condition becomes $$ \sum N_n 2^{-\epsilon 2^{N_{n-1}}} =\sum n^3 2^{(-\epsilon+1/n)2^{N_{n-1}}}< \infty $$ for all $\epsilon >0$, and this holds because the $N_n$ increase rapidly.

  • $\begingroup$ thanks a lot for this wonderfully worked out example. If I understand things correctly, then what matters here of course is that I am looking for this delta neighbourhood. I could have also asked if $\liminf_{\epsilon \to 0} \frac{\sum_n 2^{-\varepsilon y_n}}{\sum_n 2^{-\varepsilon z_n}}>0$ holds true to better quantify their relative size to one another? I am not entirely sure how the series of your constructed $z_n$ scales with $\varepsilon$ so I don't see it right away? $\endgroup$ Mar 20, 2021 at 19:11
  • 2
    $\begingroup$ @PritamBemis: I think the basic issue here is that $\sum 2^{-\epsilon u_n}$ (for $u=y$ or $u=z$) only depends on what values $u_n$ takes, not on where exactly these occur, while the other series $\sum x_nu_n$ can of course be quite sensitive to rearrangements (for given $x$). This suggests that we shouldn't really expect a very close connection between the two, and I think the situation is still essentially the same in the modification you suggested. $\endgroup$ Mar 20, 2021 at 20:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.