12

Evaluate $$\sum_{n=1}^{38}\sin\left(\frac{n^8\pi}{38}\right)$$

I have found the problem on this page.

I have no idea how to do it. Thank you very much.

Ayman Hourieh
  • 39,603
kong
  • 233
  • @AymanHourieh what do you mean by power reduction? you should notice that the power is inside not on the sin! – HorizonsMaths Jul 21 '14 at 19:06
  • 1
    The big clue on this, I think, is realizing that it's not just a trigonometry question... – Semiclassical Jul 21 '14 at 19:07
  • @Mercy Yeah, misread. Never mind. I'll delete my comment. – Ayman Hourieh Jul 21 '14 at 19:08
  • 3
    Well Mathematica simplifies this to $\sqrt{19}$. I have found two more examples of this kind, $$\sum_{n=1}^{22}\sin \frac{\pi n^8}{22}=\sqrt{11},\qquad \sum_{n=1}^{46}\sin \frac{\pi n^8}{46}=\sqrt{23},$$ but so far I don't see how to characterize the numbers for which this pattern works. Hopefully this will provide some insight to others. – Start wearing purple Jul 21 '14 at 19:32
  • @O.L. interesting. So, generally it's $\sqrt \frac N 2$ where $N$ is upper limit and the denominator? – Kaster Jul 21 '14 at 19:35
  • @Kaster I would say "sometimes" instead of "generally". It doesn't work in general even if $N/2$ is prime. – Start wearing purple Jul 21 '14 at 19:38
  • It's not restricted to $N/2$ being prime, either. It works for $N=18$ and $N=21$, for instance. – Semiclassical Jul 21 '14 at 19:47
  • @Semiclassical, yeah, I did small experiment in Mathematica, and it seems to be working sporadically. Although, I don't think it works for $N = 21$. It doesn't work for any odd $N < 100$ for that matter. – Kaster Jul 21 '14 at 19:50
  • Meant $N=2*21$, sorry. @Kaster – Semiclassical Jul 21 '14 at 19:51
  • It seems we need some congruence arithmetic. – user5402 Jul 21 '14 at 20:34
  • Agree with Semiclassical. This is NOT a problem on trigonometry but rather of elementary number theory and Gauss sums as is apparent from Jack D'Aurizio's answer (+1). What this means that if you are only armed with trig identities there is relatively little hope (except possibly in some simple special case). – Jyrki Lahtonen Jul 21 '14 at 20:48
  • This approach seems to work well: split the sum accordingly to the parity of $n$. You get the imaginary part of a Gauss sum for even $n$s, and a trigonometrical sum that vanishes if $-2$ is a fourth power $\pmod{N/2}$, assuming that $N$ is odd. – Jack D'Aurizio Jul 21 '14 at 20:49
  • I guess what I'd like to see is something a bit more group-theoretic (e.g. establishing that sums of roots of unity vanish due to the sum being invariant under a rotation.) @JackD'Aurizio – Semiclassical Jul 21 '14 at 20:51

3 Answers3

8

We have: $$\sum_{n=1}^{38}\sin\left(\frac{n^8 \pi}{38}\right) = \sum_{k=0}^{18}\sin\left((2k+1)^8 \frac{2\pi}{4\cdot 19}\right)+\sum_{k=0}^{18}\sin\left(2^6 k^8 \frac{2\pi}{19}\right),$$ where the first sum vanishes because $-2$ is a fourth power $\pmod{19}$, since $5^4+2\equiv0\pmod{19}$, while the second sum is just the imaginary part of a Gauss sum: $$\sum_{k=0}^{18}\sin\left(2^6 k^8\frac{2\pi}{19}\right)=\Im\sum_{m=0}^{18}\left(\frac{m}{q}\right)\exp\left(7\cdot\frac{2\pi i m}{19}\right)=\sqrt{19}$$ because the eighth powers $\pmod{19}$ are just the quadratic residues.

Jack D'Aurizio
  • 353,855
2

Again not a solution, but I noted that

$$ \left( \sum_{n=1}^{14} n^8 \mod (2 \times 14) \right) \mod (2 \times 14) = 7,\\ \sum_{n=1}^{14} \sin \left( \frac{\pi n^8}{14} \right) = \sqrt{7}. $$

$$ \left( \sum_{n=1}^{22} n^8 \mod (2 \times 22) \right) \mod (2 \times 22) = 11,\\ \sum_{n=1}^{22} \sin \left( \frac{\pi n^8}{22} \right) = \sqrt{11}. $$

$$ \left( \sum_{n=1}^{38} n^8 \mod (2 \times 38 ) \right) \mod (2 \times 38) = 19,\\ \sum_{n=1}^{38} \sin \left( \frac{\pi n^8}{38} \right) = \sqrt{19}. $$


And other strange things:

$$ \left( \sum_{n=1}^{10} n^2 \mod (2 \times 10) \right) \mod (2 \times 10) = 5,\\ \sum_{n=1}^{10} \sin \left( \frac{\pi n^2}{10} \right) = \sqrt{5}. $$

$$ \left( \sum_{n=1}^{14} n^2 \mod (2 \times 14) \right) \mod (2 \times 14) = 5,\\ \sum_{n=1}^{14} \sin \left( \frac{\pi n^2}{14} \right) = \sqrt{7}. $$

$$ \left( \sum_{n=1}^{22} n^2 \mod (2 \times 22) \right) \mod (2 \times 22) = 11,\\ \sum_{n=1}^{22} \sin \left( \frac{\pi n^2}{22} \right) = \sqrt{11}. $$

$$ \left( \sum_{n=1}^{26} n^2 \mod (2 \times 26) \right) \mod (2 \times 26) = 13,\\ \sum_{n=1}^{26} \sin \left( \frac{\pi n^2}{26} \right) = \sqrt{13}. $$


Let us write

$$ Q(p,q) = \left( \sum_{n=1}^{2p} n^q \mod (4p) \right) \mod (4p),\\ S(p,q) = \sum_{n=1}^{2p} \sin \left( \frac{\pi n^q}{2p} \right). $$

Then we find

$$ \begin{array}{cccc} p & q & Q(p,q) & S(p,q)\\ \hline 5 & 2 & 5 & \sqrt{5}\\ 7 & 2 & 7 & \sqrt{7}\\ 11 & 2 & 11 & \sqrt{11}\\ 13 & 2 & 13 & \sqrt{13}\\ 17 & 2 & 17 & \sqrt{17}\\ 19 & 2 & 19 & \sqrt{19}\\ 23 & 2 & \color{red}{21} & \sqrt{23}\\ \hline 7 & 4 & 7 & \sqrt{7}\\ 11 & 4 & 11 & \sqrt{11}\\ 38 & 4 & 19 & \sqrt{19}\\ \hline 7 & 8 & 7 & \sqrt{7}\\ 11 & 8 & 11 & \sqrt{11}\\ 38 & 8 & 19 & \sqrt{19}\\ \end{array} $$

Hope this gives some insight for others...

  • for an answer that isn't a solution, it sure is a long one – Jam Jul 21 '14 at 21:56
  • @oliveeuler: I just noted these things and hope that they give others insight... Sorry to bother you with my post ;) – johannesvalks Jul 21 '14 at 21:57
  • 1
    Oh no bother; I just had to scroll slightly further. Though some people on the site don't seem to like it when people give incomplete answers. – Jam Jul 21 '14 at 21:59
  • Per the answer by Jack above, it's evidently enough to check whether $-2$ is a fourth power mod $N$; if so, then $2N$ will produce an example. (though to I'll be honest and say I don't follow his proof well) – Semiclassical Jul 22 '14 at 00:12
1

Not a solution, but I think this is the direction to investigate. First, let's consider instead the sum over complex exponentials (so that this particular sum will be the imaginary part). Then Mathematica gives the sum $$\sum_{n=1}^{38} \exp\left(\frac{i\pi n^8}{38}\right)=\sqrt{19}(1+i).$$ (As noted in comments, the same holds true if $38$ is replaced by some other even integers...but it's not clear which ones.)

This suggests that there's some rather generic result which we should be seeking. To approach this, note that for every eighth power we may write $n^8=76k+r$ for some integers $k,r$ with $r\in[0,76)$. Then $$ \exp\left(\frac{i \pi n^8}{38}\right)=\exp\left(2\pi k i+\frac{2\pi i r}{38}\right)=\exp\left(\frac{i\pi r}{38}\right)$$ which is a 76th root of unity.

So I think we in part need number-theoretic results: What can be said about $n^8$ for $n\in \mathbb{Z}/76\mathbb{Z}$?

Semiclassical
  • 15,842