20
$\begingroup$

I have to show that

$$g\left(\frac{1}{2015}\right) + g\left(\frac{2}{2015}\right) +\cdots + g\left(\frac{2014}{2015}\right) $$ is an integer. Here $g(t)=\dfrac{3^t}{3^t+3^{1/2}}$.

I tried to solve it using power series, but I can not finish with any convincing argument to said that the sum is an integer. (Also the use of a computer isn't allowed.)

$\endgroup$
  • $\begingroup$ What is $3.5$ ? I mean, the number $7/2$ or the multiplication $15$ ? $\endgroup$ – user1971 Oct 11 '15 at 15:09
  • $\begingroup$ It's not an integer, if $3.5 = 7/2$. $\endgroup$ – Patrick Stevens Oct 11 '15 at 15:09
  • $\begingroup$ With a computer I get the value 2005.736 if 3.5=7/2 and 1984.4776 if 3.5=15. Just check the question again. $\endgroup$ – Kartik Oct 11 '15 at 15:19
  • $\begingroup$ @Kartik I make it 559.51119983 in the first case. Mathematica: g[t_] := 3 t/(3 t + 7/2); N[Sum[g[i/2015], {i, 1, 2014}], 50] $\endgroup$ – Patrick Stevens Oct 11 '15 at 15:21
  • $\begingroup$ @PatrickStevens Yes, sorry thats right. Bug in program. 559.511 in the first case and 178.026 in the second. $\endgroup$ – Kartik Oct 11 '15 at 15:25
45
$\begingroup$

Want $\sum_{k=1}^{n-1} g\left(\frac{k}{n}\right) $ where $n$ is odd and $g(t)=\dfrac{3^t}{3^t+3^{1/2}} $.

$g(k/n) =\dfrac{3^{k/n}}{3^{k/n}+3^{1/2}} $.

$\begin{array}\\ g(k/n)+g((n-k)/n) &=\dfrac{3^{k/n}}{3^{k/n}+3^{1/2}}+\dfrac{3^{(n-k)/n}}{3^{(n-k)/n}+3^{1/2}}\\ &=\dfrac{3^{k/n}(3^{(n-k)/n}+3^{1/2})+3^{(n-k)/n}(3^{k/n}+3^{1/2})}{(3^{k/n}+3^{1/2})(3^{(n-k)/n}+3^{1/2})}\\ &=\dfrac{(3+3^{(k/n)+(1/2)})+(3+3^{(n-k)/n+(1/2)})}{3+3^{1/2}(3^{k/n}+3^{(n-k)/n})+3}\\ &=\dfrac{6+3^{1/2}(3^{k/n}+3^{(n-k)/n})}{6+3^{1/2}(3^{k/n}+3^{(n-k)/n})}\\ &= 1\\ \end{array} $

Wow! This was completely unexpected.

Since the sum of paired terms is one, if $n$ is odd, and there are $\frac{n-1}{2}$ pairs the sum is $\frac{n-1}{2}$.

It looks like any number can be substituted for $3$ and this will work.

$\endgroup$
  • 7
    $\begingroup$ And this put me over 30k points. I am proud of this answer. $\endgroup$ – marty cohen Oct 11 '15 at 23:47
  • 5
    $\begingroup$ A whiff of Gauss... nicely done. $\endgroup$ – J. M. is a poor mathematician Oct 12 '15 at 4:47
  • $\begingroup$ Thank you. Upvoted, of course. $\endgroup$ – marty cohen Oct 12 '15 at 4:48
17
$\begingroup$

This is the answer to the first version of the question:


We want to compute:

$$ \sum_{k=1}^{2014}\frac{\frac{3k}{2015}}{\frac{3k}{2015}+\frac{7}{2}}=\sum_{k=1}^{2014}\frac{6k}{6k+5\cdot 7\cdot 13\cdot 31} $$ but that number cannot be an integer because $6\cdot 2001+5\cdot 7\cdot 13\cdot 31$ is a prime.

Anyway, the value of the LHS is about $2015\int_{0}^{1}\frac{3x}{3x+\frac{7}{2}}\,dx = 2015\left(1-\frac{7}{6}\log\frac{13}{7}\right)$.

We may also estimate the difference between our sum and $2015\left(1-\frac{7}{6}\log\frac{13}{7}\right)$ through the Hermite-Hadamard inequality, since $\frac{3x}{3x+\frac{7}{2}}$ is a concave function on $[0,1]$. That gives another way for proving that our sum is not an integer, since it gives that our sum is between $559.4$ and $559.8$.


This is the answer to the second version of the question. If $$ g(t) = \frac{3^t}{3^t+3^{1/2}} $$ we have: $$ g(t)+g(1-t) = \frac{3^t}{3^t+3^{1/2}}+\frac{3^{1-t}}{3^{1-t}+3^{1/2}}=\frac{1}{1+3^{1/2-t}}+\frac{1}{1+3^{t-1/2}}=\color{red}{1}$$ hence the claim is trivial.

$\endgroup$
  • $\begingroup$ By the way, Mathematica says that the difference between the two expressions is approx 0.2308. The sum has value approx 559.511199837. $\endgroup$ – Patrick Stevens Oct 11 '15 at 15:19
  • $\begingroup$ @PatrickStevens: $559.742$ or $559.511$, we still don't have an integer. $\endgroup$ – Jack D'Aurizio Oct 11 '15 at 15:21
  • $\begingroup$ Agreed. Just wanted to quantify "the value is about". $\endgroup$ – Patrick Stevens Oct 11 '15 at 15:22
  • 2
    $\begingroup$ nice answer! Right now I wonder, is there any fast way to see that $6\cdot 2001+5\cdot 7\cdot 13\cdot 31$ is prime? Is this actually the only prime in the denominator? $\endgroup$ – user190080 Oct 11 '15 at 15:22
  • 2
    $\begingroup$ @user190080: actually there are plenty of primes among the denominators. I just checked a large denominator for primality, since that gives in a very straightforward way that the sum cannot be an integer, like in the classical proof of $H_n\not\in\mathbb{Z}$ for any $n>1$. $\endgroup$ – Jack D'Aurizio Oct 11 '15 at 15:24
8
$\begingroup$

By a rigid translation you can see that $g$ is an odd function around the point $(\frac 1 2, \frac 1 2)$:

$$ \begin{align} f(x) &= g(x+\frac 1 2) - \frac 1 2 \\ &= \frac{3^{x + 1/2}}{3^{x + 1/2} + 3^{1/2}} - \frac 1 2 \\ &= \frac{3^x 3^{1/2}}{3^x 3^{1/2} + 3^{1/2}} - \frac 1 2 \\ &= \frac{3^x 3^{1/2}}{(3^x+1) 3^{1/2}} - \frac 1 2 \\ &= \frac{3^x}{3^x+1} - \frac 1 2 \\ &= \frac{2(3^x)-(3^x+1)}{2(3^x+1)} \\ &= \frac{1}{2}\frac{3^x-1}{3^x+1} \\ &= \frac{1}{2}\frac{e^{x\log(3)}-1}{e^{x\log(3)}+1} \\ &= c_1 \tanh(c_2x) \ \ \text{ with $c_1 = \frac 1 2$ and $c_2 = \log(3)$ } \end{align} $$

The last formula is the hyperbolic tangent of a multiple of $x$, so it's symmetric with respect to the origin. From this, it follows that the $f(x) + f(-x) = 0$, which, after the transformation, means $g(x) + g(1-x) = 1$.

From this the claim follows.

$\endgroup$
2
$\begingroup$

$$\begin{align} &g(t)=\frac {3^t}{3^t+3^{\frac 12}}=\frac {3^{t-\frac12}}{3^{t-\frac 12}+1}\\ \Rightarrow \qquad &g\left(\frac12+u\right)+g\left(\frac12-u\right)=\frac {3^u}{3^u+1}+\frac {3^{-u}}{3^{-u}+1}=\frac {3^u}{3^u+1}+\frac {1}{1+3^u}=1\\\\ \sum_{r=1}^{2014}g\left(\frac r{2015}\right) &=\sum_{r=1}^{1007}g\left(\frac r{2015}\right)+\sum_{r=1008}^{2014}g\left(\frac r{2015}\right)\\ &=\sum_{i=1}^{1007}g\left(\frac{1007\frac12-(i-\frac12)}{2015}\right)+g\left(\frac{1007\frac12+(i-\frac12)}{2015}\right)\\ &=\sum_{i=1}^{1007}g\left(\frac12-\frac{i-\frac12}{2015}\right)+g\left(\frac12+\frac{i-\frac12}{2015}\right)\\ &=\sum_{i=1}^{1007}1\\ &=1007\qquad\blacksquare\end{align}$$

$\endgroup$
2
$\begingroup$

Notice that $g (x) = g (1 - x)$. Indeed

$$g(x)+g(1-x) = \frac {1} {1 + 3^{1/2-x}} + \frac {1} {1+3^{x-1/2}} = 1$$.

Then we conclude by

$$\sum_{k = 1}^{2014} g\left(\frac{k}{2015}\right) = \sum_{k = 1}^{1007} \left ( g\left(\frac{k}{2015} \right ) + g\left(\frac{2015 - k}{2015} \right ) \right ) = 1007.$$

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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