26

We have,

$$\big(\cos(\tfrac{2\pi}{5})^{1/2}+(-\cos(\tfrac{4\pi}{5}))^{1/2}\big)^2 = \tfrac{1}{2}\left(\tfrac{-1+\sqrt{5}}{2}\right)^3\tag{1}$$

$$\big(\cos(\tfrac{2\pi}{7})^{1/3}+\cos(\tfrac{4\pi}{7})^{1/3}+\cos(\tfrac{6\pi}{7})^{1/3}\big)^3 = \frac{5-3\cdot 7^{1/3}}{2}\tag{2}$$

$$\big(\cos(\tfrac{2\pi}{11})^{1/5}+\cos(\tfrac{4\pi}{11})^{1/5}+\dots+\cos(\tfrac{10\pi}{11})^{1/5}\big)^5 = x?\tag{3}$$

Question: What degree is the minimal polynomial of $x$? Since the previous two are deg 2 and 3, I had assumed (3) would be deg 5, but Mathematica does not recognize it as a quintic with small coefficients, nor a 25th deg (even after using 500-decimal digit precision, though I am not sure of the latter result).

mathreadler
  • 25,824
  • How have you derived the identities? For cubic, we can set $y=(\frac z2)^{\frac13}$ here in my answer http://math.stackexchange.com/questions/225088/prove-frac14-sec2-frac2-pi7-frac14-sec2-frac4-pi7 and then use Vieta's formula – lab bhattacharjee Dec 19 '13 at 05:53
  • The first is easy, while the second is at the end of http://mathworld.wolfram.com/TrigonometryAnglesPi7.html. The third is unknown. – Tito Piezas III Dec 19 '13 at 16:19
  • Are you sure about the first "identity"? In particular, $\cos \frac{4\pi}{5}$ is negative, so taking square root of it results in purely imaginary number; so the sum of two terms is neither real, nor purely imaginary... and thus its square cannot be real. – Peter Košinár Dec 19 '13 at 22:40
  • Oops, sorry, it's a typo. I've fixed it now. – Tito Piezas III Dec 20 '13 at 01:56
  • All cosines in (2) are roots of a (rational) cubic equation. So one can use Ramanujan's formula for the sum of cubic roots of roots of a cubic equation and obtain (2). All cosines in (3) are of course roots of a quintic equation (namely, $32x^5+16x^4-32x^3-12x^2+6x+1=0$)... – Grigory M Dec 20 '13 at 08:04
  • 1
    @GrigoryM: What Ramanujan formula? Regarding (3), since $y^5=a_i$ has five roots, and we are to do so for five $a_i=\cos(n)$, then I think it has maximum degree $5^5 = 3125$ though it may factor since the $a_i$ are not random. I tested the form $z=(y_1+y_2\zeta^{k}+y_3\zeta^{2k}+y_4\zeta^{3k}+y_5\zeta^{4k})^5$ where the $y_i = \cos(n)^{1/5}$ and $\zeta^5=1$ and tried to form a 125-deg resolvent, but it still doesn't seem to be the minimal deg. – Tito Piezas III Dec 20 '13 at 20:03
  • (Re: What Ramanujan formula) see (3.8) and (3.18) in http://www.maa.org/sites/default/files/pdf/upload_library/22/Ford/Berndt-Bhargava644-656.pdf – Grigory M Dec 20 '13 at 20:08
  • 1
    I'm not sure there exists a (good) generalization in this direction — but maybe you'll find another generalization of the second formula of interest – Grigory M Jan 20 '14 at 08:33
  • Tito Piezas III what do you want to see in RHS ? Nested radicals ? – davidoff303 Jan 11 '16 at 11:52
  • @davidoff303: Yes, as nested radicals. As $x$ is an algebraic number of degree $n$, I'll be happy to know what is $n$. – Tito Piezas III Jan 11 '16 at 12:53
  • I would have left the negative sign off the second cosine in (1). Depending on the choice of square roots you then get $(1/2)\pm i$. – Oscar Lanzi Apr 30 '18 at 15:20

1 Answers1

5

As some people said, the 2nd formula is easy to derive. In Maple, there are commands to get the minimal polynomial of LHS of the 2nd. The following method works for the 2nd formula, but not for the 3rd.

Prove that $\sqrt[3]{\cos\frac{2\pi}{7}} + \sqrt[3]{\cos\frac{4\pi}{7}} + \sqrt[3]{\cos\frac{6\pi}{7}} = \sqrt[3]{\frac{5-3\sqrt[3]{7}}{2}}$.

$\textit{Proof}$: $\cos\frac{2\pi}{7},\cos\frac{4\pi}{7},\cos\frac{6\pi}{7}$ are the three distinct real roots of $8x^3+4x^2-4x-1=0$. Let $u_1=\sqrt[3]{\cos\frac{2\pi}{7}}, u_2=\sqrt[3]{\cos\frac{4\pi}{7}}, u_3=\sqrt[3]{\cos\frac{6\pi}{7}}$ which are all real. All roots of $8u^9+4u^6-4u^3-1=0$ are $u_1, t u_1, t^2 u_1, u_2, t u_2, t^2 u_2, u_3, t u_3, t^2 u_3$ where $t = e^{2\pi i/3}$.

Let $(u-u_1)(u-u_2)(u-u_3)=u^3+a_2u^2+a_1u - 1/2$. We have \begin{align} (u-tu_1)(u-tu_2)(u-tu_3) &=u^3+ta_2u^2+t^2a_1u - 1/2, \\ (u-t^2u_1)(u-t^2u_2)(u-t^2u_3) &=u^3+t^2a_2u^2+ta_1u - 1/2. \end{align} It follows from \begin{align} &8u^9+4u^6-4u^3-1 \\ =\ & 8(u^3+a_2u^2+a_1u - 1/2)(u^3+ta_2u^2+t^2a_1u - 1/2)(u^3+t^2a_2u^2+ta_1u - 1/2) \end{align} that \begin{align} a_2^3-3a_1a_2-2 &=0, \\ 4a_1^3+6a_1a_2+5 &= 0. \end{align} From these two equations, we get $a_1 = \frac{a_2^3-2}{3a_2}$ and $4a_2^9+30a_2^6+75a_2^3-32=0$. Let $f(v)=4v^9+30v^6+75v^3-32$. We have $f'(v)=9v^2(2v^3+5)^2\ge 0$. Thus, $f(v)=0$ has a unique real root. It is easy to check that $v=-\sqrt[3]{\frac{5-3\sqrt[3]{7}}{2}}$ is a root of $f(v)=0$. Thus, $a_2 = -\sqrt[3]{\frac{5-3\sqrt[3]{7}}{2}}$. Thus, $u_1+u_2+u_3 = -a_2 = \sqrt[3]{\frac{5-3\sqrt[3]{7}}{2}}$. This completes the proof.

In general, suppose $x_1,x_2,x_3$ are the three distinct real roots of $x^3+bx^2+cx+d^3=0$, where $b,c,d$ are rational numbers. We have \begin{equation} \sqrt[3]{x_1}+\sqrt[3]{x_2}+\sqrt[3]{x_3} = \sqrt[3] {-\frac{3}{2}\Big(\sqrt[3]{Q+4\sqrt{\Delta}}+\sqrt[3]{Q-4\sqrt{\Delta}}\Big) -b-6d }, \end{equation} where $Q = 24bd^2+36d^3+4bc+24cd, \Delta = -4b^3d^3-27d^6+18bcd^3+b^2c^2-4c^3$. Here, $\Delta$ is the discriminant of $x^3+bx^2+cx+d^3=0$.

For the 3rd, similarly, $x_k = \cos\frac{2k\pi}{11}, k=1,\cdots, 5$ are the five distinct real roots of $32x^5+16x^4-32x^3-12x^2+6x+1=0$. Let $u_1 = \sqrt[5]{\cos\frac{2\pi}{11}}, u_2= \sqrt[5]{\cos\frac{4\pi}{11}}, u_3 = \sqrt[5]{\cos\frac{6\pi}{11}}, u_4 = \sqrt[5]{\cos\frac{8\pi}{11}}, u_5= \sqrt[5]{\cos\frac{10\pi}{11}}$. All roots of $32u^{25}+16u^{20}-32u^{15}-12u^{10}+6u^5+1=0$ are $t^j u_k, k=1,2,3,4, 5; j=0,1,2,3,4$ where $t= e^{2\pi i/5}$.

Let $\prod_{k=1}^5 (u-u_k) = u^5 + a_4u^4+a_3u^3+a_2u^2+a_1u+\frac{1}{2}$. Similarly, we have four equations of \begin{align} 0 & = a_4^5-5a_3a_4^3+5a_2a_4^2+5a_3^2a_4-5a_1a_4-5a_2a_3+2, \\ 0 & = 8a_1^5-20a_1^3a_2+10a_1^2a_3+10a_1a_2^2-5a_1a_4-5a_2a_3+1,\\ 0 & = -10a_1a_2a_4^3+10a_1a_3^2a_4^2+10a_2^2a_3a_4^2-10a_2a_3^3a_4+2a_3^5+10a_1^2a_4^2-10a_1a_2a_3a_4\\ &\quad-10a_1a_3^3-10a_2^3a_4+10a_2^2a_3^2-5a_3a_4^3+10a_1^2a_3+10a_1a_2^2+10a_2a_4^2+10a_3^2a_4\\ &\quad -15a_1a_4-15a_2a_3+7,\\ 0 & = -40a_1^3a_3a_4+40a_1^2a_2^2a_4+40a_1^2a_2a_3^2-40a_1a_2^3a_3+8a_2^5-40a_1^3a_2+20a_1^2a_4^2\\ &\quad-20a_1a_2a_3a_4-20a_1a_3^3-20a_2^3a_4+20a_2^2a_3^2+40a_1^2a_3+40a_1a_2^2+10a_2a_4^2\\ &\quad +10a_3^2a_4-30a_1a_4-30a_2a_3+13. \end{align}

However, it seems no hope for determination of $a_4$.

River Li
  • 37,323
  • 2
    I don't mean to be rude, by I find this utter unreadable. Could you perhaps make use of some displayed math environments (using $$ (math here) $$) or the align environment (\begin{align} \end{align}) to make it easier to follow? You can also find other formatting tips in the MathJax tutorial. – Xander Henderson Aug 17 '18 at 03:19
  • Please copy your answer here to this link: https://math.stackexchange.com/questions/3798002/is-it-true-that-sum-i-1n-ngx-ig-gx-i-ge-n2g-g2n-for-all. In a week, I may copy your answer into the link so that the answer is preserved. The answer will be a Community Wiki post, and you can remove the part that belongs to you, and put it in your own separate answer. (I'm not sure if you were notified from my comment in the deleted post.) – Batominovski Aug 21 '20 at 14:30
  • @Batominovski Thanks. I saw your comment there. But I want to make more progress for that problem (or even prove it) before posting as a new answer. – River Li Aug 21 '20 at 14:43
  • @RiverLi Can you please confirm that you got a notification? I never did this nor ever received a message from deleted posts. Did you see a "red notification" when I made a comment in your deleted post? – Batominovski Aug 21 '20 at 14:45
  • @Batominovski I did not get a notification from the deleted post (although you added a comment on my answer there). But I got notifications when you added comments here. – River Li Aug 21 '20 at 14:51
  • @RiverLi Oh, thanks. I will keep this in mind for future "copy" request. – Batominovski Aug 21 '20 at 14:52
  • @Batominovski Thank you for saving this interesting question. – River Li Aug 21 '20 at 14:53
  • @RiverLi It's a shame that there are people who are determined to delete interesting contents. Since you specifically told me that you wanted to spend some time with the question, I will not copy your answer even after a week has passed, to honor your wish. But my question is unlikely to be closed or deleted (as they cannot close it for being a duplicate, since the original has been deleted, and the new question met the requirement for being a "good question"). So, it will be there when you want to answer it. – Batominovski Aug 21 '20 at 14:56
  • I agree with you. Thanks. – River Li Aug 21 '20 at 15:08