4
$\begingroup$

In two previous problems (Closed expression for sum $\sum_{k=1}^{\infty} (-1)^{k+1}\frac{\left\lfloor \sqrt{k}\right\rfloor}{k}$ and Closed expression for sum $\sum_{k = 1}^{\infty} \frac{\left\lfloor \sqrt{k} \right \rfloor}{k^2}$) the infinite sums contained the floor function (of a square root) in the numerator.

Here we ask, in a simple example, what happens if the floor function is in the denominator.

Question: what is the closed form of $\sum _{k=1}^{\infty } \frac{(-1)^{k+1}}{\left\lfloor \sqrt{k}\right\rfloor }$

$\endgroup$
3
  • $\begingroup$ It turns out this is the same for $\sqrt[m]{k}$ for any positive integer $m$. See my generalization here: math.stackexchange.com/questions/3456631/… $\endgroup$ Commented Nov 30, 2019 at 5:01
  • $\begingroup$ Dear downvoters, I would appreciate some information on your motivation to downvote my question. I suppose you took a responsible decision which means that you took into account what downvoting officially means, quote: "This question does not show any research effort. It is unclear and not useful." My comments: (1) In this case I had to conceil the solution steps in order not to spoil the simple and surprising result. Therefore I did not ask if there's a closed form but how it looks like. My research effort in closely related problems can be found in and via the reference I gave. $\endgroup$ Commented Dec 1, 2019 at 20:02
  • $\begingroup$ Dear downvoters ctd. (2) The question is clearly indicated by bold letters in the text and the content should be clear (the type of question is common and abundant in this forum). (3) Whether the question is useful or not is to a high degreee a matter of taste. At least one upvoter, one generalizer and I considered it useful. Finally, I deplore missing consistency. My question was downvoted today 4 times by -2. Nevertheless this "bad" question has given the opportunity to gain of 7 upvotes in one answer (which I consider completely justified). $\endgroup$ Commented Dec 1, 2019 at 20:03

2 Answers 2

12
$\begingroup$

For any $k \geq 1$, we can uniquely write $k=n^2+\ell$, with $0\leq \ell \leq 2n$. Then, $\lfloor \sqrt{k}\rfloor = n$, so that $$\begin{align} \sum_{k=1}^\infty \frac{(-1)^{k+1}}{\lfloor \sqrt{k}\rfloor} &= \sum_{n=1}^\infty\sum_{\ell=0}^{2n} \frac{(-1)^{n^2+\ell+1}}{n} = -\sum_{n=1}^\infty \frac{(-1)^{n^2}}{n} \sum_{\ell=0}^{2n} (-1)^{\ell}\\ &= -\sum_{n=1}^\infty \frac{(-1)^{n}}{n} \sum_{\ell=0}^{2n} (-1)^{\ell} = -\sum_{n=1}^\infty \frac{(-1)^{n}}{n} \end{align}$$ since $(-1)^{n^2}=(-1)^n$ and $\sum_{\ell=0}^{2n} (-1)^{\ell}=1$ for all $n$.

It follows that $$ \sum_{k=1}^\infty \frac{(-1)^{k+1}}{\lfloor \sqrt{k}\rfloor} = -\sum_{n=1}^\infty \frac{(-1)^{n}}{n} = \boxed{\log 2} $$

$\endgroup$
7
  • 1
    $\begingroup$ Very good solution. I found it interesting that effectively the floor function "erases" the square root $\endgroup$ Commented Nov 30, 2019 at 1:01
  • $\begingroup$ Yes, it's a neat coincidence... $\endgroup$
    – Clement C.
    Commented Nov 30, 2019 at 1:02
  • $\begingroup$ Can you also find $\sum_{k=1}^\infty \frac{x^k}{\lfloor \sqrt{k}\rfloor}$ ? $\endgroup$ Commented Nov 30, 2019 at 1:06
  • $\begingroup$ That's not going to be as easy. $x^{n^2}\neq x^n$, for a start. $\endgroup$
    – Clement C.
    Commented Nov 30, 2019 at 1:07
  • 1
    $\begingroup$ @Dr.WolfgangHintze But that's hardly a closed form, is it? $\endgroup$
    – Clement C.
    Commented Nov 30, 2019 at 1:50
3
$\begingroup$

OK. Here's my generalization which ran into some resistance when posted as a separate problem.

Let $f(x)$ be such that $f(1) = 1, f'(x) > 0, f''(x) < 0, f(x) \to \infty, n \in \mathbb{N} \implies f^{(-1)}(n)\in \mathbb{N} $.

($f^{(-1)}(n)$ is the inverse function of $f$)

What can we say about $$S=\sum_{k=1}^{\infty} \dfrac{(-1)^{k+1}}{\lfloor f(k) \rfloor} $$ Let $g$ be the inverse function of $f$, so $f(g(x)) = g(f(x)) = x $.

Let $u(n) = \begin{cases} 0 \text{ if } n \text{ odd}\\ 1 \text{ if } n \text{ even}\\ \end{cases} =\dfrac{(-1)^n+1}{2}. $

\begin{align} S &=\sum_{k=1}^{\infty} \dfrac{(-1)^{k+1}}{\lfloor f(k) \rfloor}\\ &=\sum_{n=1}^{\infty} \sum_{k=g(n)}^{g(n+1)-1} \dfrac{(-1)^{k+1}}{\lfloor f(k) \rfloor}\\ &=\sum_{n=1}^{\infty} \sum_{k=g(n)}^{g(n+1)-1} \dfrac{(-1)^{k+1}}{\lfloor n \rfloor}\\ &=\sum_{n=1}^{\infty} \dfrac1{n}\sum_{k=g(n)}^{g(n+1)-1} (-1)^{k+1}\\ &=\sum_{n=1}^{\infty} \dfrac1{n}\sum_{k=0}^{g(n+1)-g(n)-1} (-1)^{k+g(n)+1}\\ &=\sum_{n=1}^{\infty} \dfrac{(-1)^{g(n)+1}}{n}\sum_{k=0}^{g(n+1)-g(n)-1} (-1)^{k}\\ &=\sum_{n=1}^{\infty} \dfrac{(-1)^{g(n)+1}}{n}u(g(n+1)-g(n)-1)\\ \end{align}

If $f(k) = \sqrt{k}$, then $g(n) = n^2$ so $u(g(n+1)-g(n)-1) =u(2n) =1 $ and $(-1)^{g(n)+1} =(-1)^{n^2+1} =(-1)^{n+1} $ so $$S =\sum_{n=1}^{\infty} \dfrac{(-1)^{g(n)+1}}{n}u(g(n+1)-g(n)-1) =\sum_{n=1}^{\infty} \dfrac{(-1)^{n+1}}{n} =\ln(2). $$

If $f(k) = \sqrt[3]{k}$, then $g(n) = n^3$ so $u(g(n+1)-g(n)-1) =u(3n^2+3n) =u(3n(n+1)) =1 $ and $(-1)^{g(n)+1} =(-1)^{n^3+1} =(-1)^{n+1} $ so $$S =\sum_{n=1}^{\infty} \dfrac{(-1)^{g(n)+1}}{n}u(g(n+1)-g(n)-1) =\sum_{n=1}^{\infty} \dfrac{(-1)^{n+1}}{n} =\ln(2). $$

If $f(k) = \sqrt[m]{k}$, then $g(n) = n^m$ so

$\begin{array}\\ u(g(n+1)-g(n)-1) &=u((n+1)^m-n^m-1)\\ &=u(\sum_{j=1}^{m-1} \binom{m}{j}n^j)\\ &=u(\sum_{j=1}^{\lfloor \frac{m-1}{2} \rfloor} (\binom{m}{j}n^j+\binom{m}{m-j}n^{m-j}) \qquad\text{central binomial coefficient is even}\\ &=u(\sum_{j=1}^{\lfloor \frac{m-1}{2} \rfloor} (\binom{m}{j}(n^j+n^{m-j}))\\ &=u(\sum_{j=1}^{\lfloor \frac{m-1}{2} \rfloor} (\binom{m}{j}n^j(1+n^{m-2j}))\\ &=1 \qquad\text{since }n^j(1+n^{m-2j}) \text{ is even}\\ \end{array} $

and $(-1)^{g(n)+1} =(-1)^{n^m+1} =(-1)^{n+1} $ so $$S =\sum_{n=1}^{\infty} \dfrac{(-1)^{g(n)+1}}{n}u(g(n+1)-g(n)-1) =\sum_{n=1}^{\infty} \dfrac{(-1)^{n+1}}{n} =\ln(2) $$

$\endgroup$
2
  • $\begingroup$ @ Marty Cohen +1 Nicely done. Have you seen my last question math.stackexchange.com/q/3457629/198592 ? The minor change makes the problem much more complicated since, in your terms, the condition $n \in \mathbb{N} \implies f^{(-1)}(n)\in \mathbb{N} $ is voilated. I think it would be a real challenge to explore the limits of the methods we have seen so far. But there is "resistance" as well ... $\endgroup$ Commented Dec 1, 2019 at 9:58
  • $\begingroup$ $u((n+1)^m-n^m-1)=1$ is obviously, because $(n+1)^m-n^m-1$ is even. (If $n$ is even, then $n+1$ is odd, or if $n$ is odd, then $n+1$ is even. In both case, the number $(n+1)^m-n^m-1$ is even.) $\endgroup$
    – Riemann
    Commented Jun 10, 2020 at 8:33

You must log in to answer this question.

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