4
$\begingroup$

I came across the following inequality, which should hold for any integer $k\geq 1$:

$$\sum_{j=0}^{k-1}\frac{(-1)^{j}2^{k-1-j}\binom{k}{j}(k-j)}{2k+1-j}\leq \frac{1}{3}.$$

I have been struggling with this statement for a while. It looks valid for small $k$, but a formal proof seems out of reach with my tools. Any suggestions on how to approach this?

$\endgroup$
2
  • $\begingroup$ Thank you for the remark, @Carlo Beenakker. Note that the original statement does not hold in the continuous sense. E.g. if we substitute $k=\frac{3}{2}$, then the value is $\frac{3}{4\sqrt{3}}>1/2$. $\endgroup$
    – macat
    Mar 2, 2022 at 19:38
  • $\begingroup$ my mistake, apologies. $\endgroup$ Mar 2, 2022 at 19:49

2 Answers 2

5
$\begingroup$

For $j=0,\dots,k-1$, \begin{equation*} \frac1{2k+1-j}=\int_0^1 x^{2k-j}\,dx. \end{equation*} So, \begin{equation*} \begin{aligned} s:=&\sum_{j=0}^{k-1}\frac{(-1)^{j}2^{k-1-j}\binom{k}{j}(k-j)}{2k+1-j} \\ &=\int_0^1 dx\,\sum_{j=0}^{k-1}(-1)^{j}2^{k-1-j}\binom{k}{j}(k-j)x^{2k-j} \\ &=\int_0^1 dx\,kx^{k+1}(2x-1)^{k-1}=I_1+I_2, \end{aligned} \tag{1}\label{1} \end{equation*} where \begin{equation*} I_1:=\int_0^{1/2} dx\,kx^{k+1}(2x-1)^{k-1}=(-1)^{k-1}\frac{k!(k+1)!}{(2k+1)!}\le \frac1{2^{2k+2}}\le\frac1{16k^2}, \tag{2}\label{2} \end{equation*} \begin{equation*} I_2:=\int_{1/2}^1 dx\,ke^{g(x)}, \end{equation*} \begin{equation*} g(x):=(k+1)\ln x+(k-1)\ln(2x-1). \end{equation*} Next, $g(1)=0$, $g'(1)=3k-1$, and, for $x\in(1/2,1)$, \begin{equation*} g''(x)=-\frac{k+1}{x^2}-\frac{4(k-1)}{(2x-1)^2}\le-(k+1)-4(k-1)=3-5k \end{equation*} and hence $g(x)\le h(x):= (3k-1)(x-1)+(3-5k)(x-1)^2/2$. So, \begin{equation*} I_2\le\int_{-\infty}^1 dx\,ke^{h(x)}=J(k):=\sqrt{\frac{\pi }{2}} e^{\frac{(1-3 k)^2}{10 k-6}} k\, \frac{\text{erf}\left(\frac{1-3 k}{\sqrt{10 k-6}}\right)+1}{\sqrt{5 k-3}}. \tag{3}\label{3} \end{equation*} Let \begin{equation*} H(k):=\text{erf}\left(\frac{1-3 k}{\sqrt{10 k-6}}\right)+1 -\left(\frac{1}{3}-\frac{1}{16k^2}\right)\frac{\sqrt{\frac{2}{\pi }} e^{-\frac{(1-3 k)^2}{10 k-6}} \sqrt{5 k-3} }{k}. \end{equation*} Then \begin{equation*} H'(k)=\frac{e^{-\frac{(1-3 k)^2}{10 k-6}} \left(160 k^4-647 k^3+75 k^2+456 k-162\right)}{48 \sqrt{2 \pi } k^4 (5 k-3)^{3/2}}>0 \end{equation*} for $k\ge4$ and $H(k)\to0$ as $k\to\infty$. So, for $k\ge4$ we have $H(k)<0$ or, equivalently, \begin{equation*} J(k)<\frac{1}{3}-\frac{1}{16k^2}. \end{equation*} Therefore and in view of \eqref{1}, \eqref{2}, and \eqref{3}, for $k\ge4$ we have \begin{equation*} s<\frac13, \end{equation*} as desired. Checking the latter inequality for $k=1,2,3$ is straightforward.

$\endgroup$
1
  • $\begingroup$ Thank you for this detailed explanation. I could not see this kind of proof coming, so thanks again! $\endgroup$
    – macat
    Mar 9, 2022 at 8:58
3
$\begingroup$

If the value of the sum is $\frac13-\varDelta(k)$, then it appears that $\varDelta(k)$ satisfies the recurrence $$ (8k+4)\varDelta(k) = (7k-5)\varDelta(k-1) + k\varDelta(k-2).$$ Note that I didn't prove this, though I assume that standard methods for deriving recurrences will suffice. I checked 500 terms. The nonnegativity of $\varDelta(k)$ of course follows trivially from the recurrence and the first few values.

$\endgroup$
2
  • $\begingroup$ I had thought about such an approach, but then decided to proceed with what seemed more readily available. :-) $\endgroup$ Mar 3, 2022 at 14:43
  • $\begingroup$ Nice observation. This could give an elementary argument --- however, showing that the recursion holds does not seem to be much fun. I will let you know if I can work out the details. Thank you for pointing this out! $\endgroup$
    – macat
    Mar 9, 2022 at 9:03

Your Answer

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

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