3
$\begingroup$

Playing with the Weierstrass $\wp$-function of the hexagonal (or triangular) lattice $\mathbb{T}$, $$ \wp'(z)^2 = 4 \wp(z)^3 - 1, $$ I noticed that the zeros of $\wp'(z) + \sqrt{3}$ are $$ \frac{\varpi}{3} + \omega, \quad e^{\pi i/3} \frac{2\varpi}{3} + \omega, \quad e^{2\pi i/3} \frac{\varpi}{3} + \omega, \qquad \omega \in \mathbb{T} $$ (clearly they are all simple), where $\varpi$ is the real period of $\wp$ given by $$ \varpi = \int_{4^{-1/3}}^{\infty} \frac{dx}{\sqrt{x^3 - 1/4}} = \frac{1}{2 \pi}\, \Gamma(1/3)^3 $$ (the zeros of $\wp'(z) - \sqrt{3}$ can be also determined). Since $$ \wp'(z)^2 - 3 = 4 \left[\wp(z)^3 - 1\right] $$ and $\wp(x)$ is real for $0 < x < \varpi$, we also get $$ \wp\left(\pm\frac{\varpi}{3}\right) = 1 $$ or, equivalently, $$ \int_1^{\infty} \frac{dx}{\sqrt{4x^3 - 1}} = \int_{\wp\left(\frac{\varpi}{3}\right)}^{\infty} \frac{dx}{\sqrt{4x^3 - 1}} = \frac{\varpi}{3} = \frac{1}{6 \pi}\, \Gamma(1/3)^3 $$ (the detailed calculations can be found in arXiv:2105.04307).

Since I am not an expert in elliptic functions, my question is whether the above facts are known.

$\endgroup$
1
  • $\begingroup$ For the record, I deleted an answer as I misunderstood the question. $\endgroup$
    – Hao Chen
    Aug 27, 2021 at 17:19

1 Answer 1

1
$\begingroup$

There is also a more algebraic perspective.

For this, it is easier (at least for me) to ``reverse engineer'' things by starting with your points, and then deducing something about the elliptic function values.

I will working on the elliptic curve $E: y^2 = 4 x^3 - 1$, which is identified with the quotient $\mathbb C/\mathbb T$ via $z \bmod \mathbb T \mapsto \bigl( \wp(z), \wp'(z) \bigr).$

Let $P$ be the point of $E$ which is the image of $\varpi/3$. It is a point of order $3$ in $E$. Now $E$ has ``complex multiplication'' by the ring $\mathcal O$ obtained by adjoining $\zeta = e^{2\pi i/3}$ to $\mathbb Z$ (because the lattice $\mathbb T$ is a submodule of $\mathbb C$ with respect to this ring). The element $\zeta$ sends $(x,y)$ to $(\zeta x, y)$.

In the ring $\mathcal O$, the number $3$ is not prime: it factors as $3 = (1 - \zeta)(1 - \zeta^{-1}) = (1-\zeta)^2 \cdot (-\zeta^{-1}).$

So among the $3$-torsion points, there are some even more special points, namely those killed by multiplication by $(1-\zeta)$ (and not just by $(1-\zeta)^2$). Looking at the formula for multiplication by $\zeta,$ we see these are the points with $x = 0$, so the points $(0,\pm i).$

There are $9$ points of order dividing $3$ altogether, namely the origin (i.e. the point at infinity), these $2$ points $(0,\pm i)$, and then the remaining $6$ points $(x,y)$ which are killed by $3$, but not by $(1-\zeta)$.

If $P$ is one of these latter points, then we see that $[1-\zeta] P = (0,\pm i).$

We can compute the $x$-coordinate of $[1-\zeta] P$ explicitly, in terms of the $x$-coordinate of $P$, just using the usual addition formula. The answer is $$x^3 - 1 = 0.$$

So we find that these $6$ points satisfy $x^3 = 1$, and then (from the equation for $E$) $y^2 = 3$.

In terms of the lattice, these points are the cosets of $\varpi/3,$ $2 \varpi/3$, $\zeta \varpi/3$, $\zeta 2 \varpi/3$, $\zeta^{-1}\varpi/3,$ and $\zeta^{-1} 2 \varpi/3$. (The remaining cosets of order $3$ are exactly $0,$ $(1 + \zeta) \varpi/3$, and $2(1+\zeta)\varpi/3$, and when you multiply these by $(1-\zeta)$, you actually land in $\mathbb T$.)

So there are $6$ lattice cosets, and $6$ sets of $(x,y)$-coordinates, which match when you apply $z \bmod \mathbb T \mapsto \bigl(\wp(z),\wp'(z)\bigr).$

The cosets $\varpi/3$, $\zeta \varpi/3$, and $\zeta^{-1} \varpi/3$ are related by successive multiplications by $\zeta,$ and so have the same $y$-value. (Either $\sqrt{3}$ or $-\sqrt{3}$.)

Similarly the cosets $2\varpi/3$, $\zeta 2 \varpi/3,$ and $\zeta^{-1} 2 \varpi/3$ have the same $y$-value. (Again, either $\sqrt{3}$ or $-\sqrt{3}$.)

To see that $\varpi/3$ actually maps to $-\sqrt{3}$, it it just a matter of checking that $\wp'(\varpi/3)$ is negative, which I'd guess one can fairly easily do.


This answer may look elaborate, but I've tried to include details to help non-experts in this perspective. I think it's fair to say that typically all such identities are part of the general theory of complex multiplication, and derivable by this kind of analysis.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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