6

I. Cubic

In 2013, I asked a question regarding an identity by Ramanujan of form,

$$\sqrt[3]{0+2\cos\tfrac{2\pi}{7}}+\sqrt[3]{0+2\cos\tfrac{4\pi}{7}}+\sqrt[3]{0+2\cos\tfrac{6\pi}{7}} = \sqrt[3]{+5-3\,\sqrt[3]{7}}$$

which later turned out to have a hidden partner,

$$\sqrt[3]{1+2\cos\tfrac{2\pi}{7}}+\sqrt[3]{1+2\cos\tfrac{4\pi}{7}}+\sqrt[3]{1+2\cos\tfrac{6\pi}{7}} = \sqrt[3]{-4+3\,\sqrt[3]{7}}$$

There are infinitely many pairs of identities $\sqrt[3]{x_1} + \sqrt[3]{x_2} + \sqrt[3]{x_3}\,$ where $x_i$ are roots of cubics. The natural question is if this has a deg-$5$ version, starting with the minimal polynomial of $2\cos\frac{2\pi}{11}$.


II. Quintic

But the analogy only goes so far: it turns out the cosine expressions under the radical in the LHS now involve products which explains why it was not as easy to find. Let, \begin{align}a &= 2\cos \frac{2\pi}{11},\quad b = 2\cos \frac{4\pi}{11},\quad c = 2\cos \frac{6\pi}{11},\\ d &= 2\cos \frac{8\pi}{11},\quad e= 2\cos \frac{10\pi}{11}\end{align}

We finally have the analogous identity,

$$\sqrt[5]{a\,b^2e^2}+\sqrt[5]{a^2bd^2}+\sqrt[5]{cd^2e^2}+\sqrt[5]{b^2c^2d}+\sqrt[5]{a^2c^2e} = \sqrt[5]{1-5\big(55+21\sqrt[5]{11}-13\sqrt[5]{11^2}-13\sqrt[5]{11^3}\big)}$$

The minpoly of $(a,b,c,d,e)$ is the well-known,

$$u^5 + u^4 - 4u^3 - 3u^2 + 3u + 1 = 0$$

while that of $x_1=ab^2e^2$ (and its conjugates) is the lesser known,

$$x^5 - 6x^4 - x^3 + 32x^2 + 16x + 1 = 0$$

So equivalently,

$$\sqrt[5]{x_1}+\sqrt[5]{x_2}+\sqrt[5]{x_3}+\sqrt[5]{x_4}+\sqrt[5]{x_5} \\ = \sqrt[5]{1-5\big(55+21\sqrt[5]{11}-13\sqrt[5]{11^2}-13\sqrt[5]{11^3}\big)}$$

The quintic in $x$ is the same given by Noam Elkies (up to sign) in this 2016 MO comment which, after some trial and error, I figured out has factorable roots. And since the quintic in $u$ is a member of the Emma Lehmer family, then by reversing this factorization, this leads to the question.


III. Transformation

The Emma Lehmer quintic is,

$$u^5 + n^2u^4 - (2n^3 + 6n^2 + 10n + 10)u^3 + (n^4 + 5n^3 + 11n^2 + 15n + 5)u^2 + (n^3 + 4n^2 + 10n + 10)u +1=0$$

The case $u=2\cos\frac{2\pi}{11}$ is just $n=-1$. Let $x=u_1u_2^2u_5^2$ for the appropriate $u_k$, then we get the new quintic,

$$\frac{(x+1)^5}{(n^4 + 5n^3 + 15n^2 + 25n + 25)} = (n^3 + 3n^2 + 4n + 3) x^4 - (n^4 + 6n^3 + 16n^2 + 23n + 11) x^3 - (n^3 + 5n^2 + 13n + 11) x^2 - (n + 2)x\\$$

The discriminant of both involve the factor $D = (n^4 + 5n^3 + 15n^2 + 25n + 25)^4$. Examples:

Let $n=1, D=71^4$, so $x^5 - 776x^4 + 4057x^3 + 2140x^2 + 218x + 1 = 0$, thus,

$$\sqrt[5]{x_1}+\sqrt[5]{x_2}+\sqrt[5]{x_3}+\sqrt[5]{x_4}+\sqrt[5]{x_5} = \sqrt[5]{1-5\big(185 - 113 \sqrt[5]{71} - 65 \sqrt[5]{71^2} - \sqrt[5]{71^3} + 12\sqrt[5]{71^4}\big)}$$

Let $n=2, D=191^4$, so $x^5 - 5916x^4 + 35345x^3 + 12425x^2 + 769x + 1 = 0$, thus,

$$\sqrt[5]{x_1}+\sqrt[5]{x_2}+\sqrt[5]{x_3}+\sqrt[5]{x_4}+\sqrt[5]{x_5} = \sqrt[5]{1-5\big({-299} - 546 \sqrt[5]{191} - 124 \sqrt[5]{191^2} + 20 \sqrt[5]{191^3} + 21 \sqrt[5]{191^4}\big)}$$

and so on.


IV. Questions

  1. Using the roots $x_i$, how do we prove that, $$\sqrt[5]{x_1}+\sqrt[5]{x_2}+\sqrt[5]{x_3}+\sqrt[5]{x_4}+\sqrt[5]{x_5} = \sqrt[5]{\beta}$$ where $\beta$ is just a quintic for general $n$? And how to show $\beta$ has the simple form, $$\beta = c_0+c_1f+c_2f^2+c_3f^3+c_4f^4$$ where $f = (n^4 + 5n^3 + 15n^2 + 25n + 25)^{1/5}$ and $c_i$ are just rationals?

  2. Since $f$ is actually $f = \pm D^{1/20}$ where we used the positive case, does the identity have a partner which uses the negative case (like in the cubic version)?


V. Postscript

With another combination of the Lehmer roots described in this MO post, then,

$$\sqrt[5]{y_1}+\sqrt[5]{y_2}+\sqrt[5]{y_3}+\sqrt[5]{y_4}+\sqrt[5]{y_5} = \color{red}0$$

analogous to orangeskid's cubic version in this MSE post.

  • Great question! What would be the property of a quintic for which the sum $\sum x_k^{\frac{1}{5}}$ is $x^{\frac{1}{5}}$ for the unique real root of another quintic? Is there a pattern that you noticed? – orangeskid Jul 19 '23 at 22:03
  • 1
    @orangeskid You know, I thinking of the exact same question: what is the necessary property of a quintic such that its roots behave in the manner as above? Elkies only gave me one example, but he must have used first principles from Galois theory to come up with it. Also, cubics were easier since they were always solvable, but that is one thing to consider: must the quintic be solvable? Second, will a one-real root quintic work as well? Third, I have a feeling these identities come in pairs because of $\pm \sqrt{d}$. So many questions. – Tito Piezas III Jul 20 '23 at 05:17

0 Answers0