(Edit: I need to revise this question with my original intent. Pls do not answer it yet. Thanks.)

Given the regular $n$-gon formed by the $n$-th roots of unity. For some $n$, how do we find $\sqrt{n}$ using the sum/difference of line segments?


It is enough to use one line segment: If $x^5=1$, then it can be the distance between the root $x_0$ on the real line, and $x_2$ in the second quadrant,

$\hskip2.2in$ enter image description here



I observed that using the sum/difference of four line segments would do. Define,

$$L(\alpha,\beta)=\sqrt{\left(\cos\big(\tfrac{2\pi\,\alpha}{17}\big)+\cos\big(\tfrac{2\pi \,\beta}{17}\big)\right)^2+\left(\sin\big(\tfrac{2\pi\,\alpha}{17}\big)-\sin\big(\tfrac{2\pi \,\beta}{17}\big)\right)^2}$$






  1. Is there an alternative to $(2)$ that is purely a sum of positive values?
  2. How do we find $(3)$? (I assume it needs $64$ line segments.)
  • $\begingroup$ So to find $(a+b\sqrt{n})/c$ for some integers $a,b,c$ is enough? $\endgroup$ – coffeemath Feb 8 '16 at 18:42
  • $\begingroup$ Yes. Minor tweaking can then isolate $\sqrt{n}$. $\endgroup$ – Tito Piezas III Feb 8 '16 at 18:43
  • $\begingroup$ @ccorn: Thanks for $n=5$. I've also revised $n=17$ since the distance formula must take into account if points are in different quadrants. $\endgroup$ – Tito Piezas III Feb 9 '16 at 5:37
  • $\begingroup$ Deleting my answer till the dust has settled. $\endgroup$ – ccorn Feb 9 '16 at 6:47
  • 1
    $\begingroup$ @coffeemath: the squareroot in $(1)$ yields $(\sqrt{5}-1)/2$. The initial $1+$ was for consistency with earlier versions of $(2)$, cf. the edit history. The actual problem (at least description-wise) is the $1+$ within the squareroot, as that suggests a distance to $-1$ rather than $+1$. $\endgroup$ – ccorn Feb 9 '16 at 7:43

If $p\equiv1\bmod4$, then $\sum_0^{p-1}e^{2\pi ia^2/p}=\sqrt p$.

And if $p≡3$ mod $4$ the right side of the sum equality is just $i\sqrt{p}$. Such sums, for either caae, are called Gauss sums.

For example in a regular hendecagon with vertices numbered 0 through 10 in rotational order, the Gauss sum shows that the distances from vertex $k$ to $11−k$, with $k$ nonzero and a negative sign attached for $k=2$, gives $\sqrt{11}$ times the circumradius.

  • $\begingroup$ Gerry: My apologies, but I asked that my question to be temporarily closed until I clarify some things. $\endgroup$ – Tito Piezas III Feb 9 '16 at 19:20
  • $\begingroup$ @TitoPiezasIII: Closing does not give you a fool-proof peace. Some eager beavers may reopen it! I could close and lock it, but then you couldn't edit the question either. Why don't you write some kind of a statement about your intent on the first line of the post, and remove that when you are done editing. That would be simpler. Not guaranteed to work, but nothing is. $\endgroup$ – Jyrki Lahtonen Feb 9 '16 at 19:26
  • $\begingroup$ @JyrkiLahtonen: Ok, I'll do that. Thanks. $\endgroup$ – Tito Piezas III Feb 9 '16 at 19:30
  • $\begingroup$ For what it's worth, Tito, a line segment is just a difference of two roots of unity. $\endgroup$ – Gerry Myerson Feb 9 '16 at 21:53

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.