16

What is special about $x^3- 6 x^2 + 4 x -2$? The 24th power of the real root - 24 is curiously close to two other numbers, one being the Ramanujan constant.

root 163

There are more of these polynomials associated with the Heegner numbers.

Heegner Polynomials

$poly_1 = x^2 -2$. The first root space has the Pythagoras constant, Silver ratio, vertices of an octagon, A4 paper, Ammann tile and the curiosity below, with numbers representing powers of $\sqrt2$.

sqrt2 powers distance wheel

$poly_2 = x^3 - 2x^2 +2x-2$. The second root space is part of tribonacci space $T = t^3 - x^2 -x-1$, notable for the snub cube. Roots $T_n$ as ${T_n}^2 - T_n$ are the roots of $poly_2$. The polynomial roots $1+2 T_n -{T_n}^2$, also in the same root space, become de Weger's example, the second best known algebraic solution for the ABC Conjecture.

snub cube

$poly_3 = x^3 - 2x-2$. This root space builds the 12 point Heilbronn solution and the 12 disk covering solution. With $r$ as the real root, the circles have radius $\sqrt{(1,r,r^2)\cdot(-3,0,1)}$, with two centers on the $x$-axis at $\sqrt{(1,r,r^2)\cdot(-7,4,0)}$ and $\sqrt{(1,r,r^2)\cdot(-1,2,-1)}$.

two twelves

Can anyone find amazing properties for the root spaces of the last three polynomials?

These are related to New Substitution Tilings Using 2, φ, ψ, χ, ρ.

Zsbán Ambrus
  • 320
  • 3
  • 18
Ed Pegg
  • 22,188
  • 1
    How does this anyway need to be placed in the combinatorics tag And by the way congrats for the 8k. Because the last (+1) is from me :-) – Rohan Shinde Jan 17 '18 at 19:13
  • 1
    Undoubtedly you know all this. Just listing a few observations: The largest roots of the last three cubics are Pisot numbers explaining why their high powers are closer and closer to integers. Also the discriminants of those cubics have the corresponding Heegner number as a factor. Of course, that doesn't explain why the 24th power would be related to the Ramanujan constant. – Jyrki Lahtonen Jan 12 '20 at 06:47
  • 2
    Some remarks: (1) If $L$ is the splitting field of $x^3-6x^2+4x-2$ then $\text{Gal}(L/\mathbb{Q})\cong S_3$ and $L$ contains $\mathbb{Q}(\sqrt{-163})$. (2) $L$ and all of its subfields have class number 1. (3) 24th powers have a rather special role in the land of modular forms (e.g. Dedekind eta function which is closely related to the j function).

    These three observations are enough to convince me that it's not just a coincidence of finding a P.V. number whose powers happen to come close to the Ramanujan constant.

    – Thomas Browning Jan 21 '20 at 03:13
  • 1
    The real root of the cubic is $$x=2+4 \sqrt{\frac{2}{3}} \cosh \left(\frac{1}{3} \cosh ^{-1}\left(\frac{15}{16}\sqrt{\frac{3}{2}}\right)\right)$$ – Claude Leibovici Oct 31 '21 at 08:02
  • 1
    OP has also posted this, on Brillhart's cubic $x^3-8x-10$ and $e^{\pi\sqrt{163}}$: https://community.wolfram.com/groups/-/m/t/2235113 – is there a simple relation between the two cubics? – Gerry Myerson Feb 18 '22 at 01:37
  • 2
    @GerryMyerson Yes: $(x-2)^3-8(x-2)-10=x^3-6x^2+4x-2$ – Thomas Browning Mar 14 '22 at 23:08

2 Answers2

9

Here is how to derive the approximation $e^{\pi\sqrt{163}}\approx x^{24}-24$.

First recall how the approximation $e^{\pi\sqrt{163}}\approx640320^3+744$ is proved. The $j$ function has $q$-expansion $$j(\tau)=\frac{1}{q}+744+196884q+21493760q^2+\cdots.$$ If we set $\tau=\frac{1}{2}(1+\sqrt{-163})$ and $q=e^{2\pi i\tau}=-e^{-\pi\sqrt{163}}$, then we have $$-640320^3=j(\tau)=-e^{\pi\sqrt{163}}+744-\frac{196884}{e^{\pi\sqrt{163}}}+\frac{21493760}{e^{2\pi\sqrt{163}}}+\cdots.$$ This gives the approximation $$e^{\pi\sqrt{163}}\approx640320^3+744.$$

For the approximation $e^{\pi\sqrt{163}}\approx x^{24}-24$, we will use the Weber modular function $$\mathfrak f(\tau)=q^{-1/48}\prod_{n=1}^\infty(1+q^{n-1/2}).$$ The $24$th power of $\mathfrak f(\tau)$ has $q$-expansion $$\mathfrak f(\tau)^{24}=q^{-1/2}\prod_{n=1}^\infty(1+q^{n-1/2})^{24}=q^{-1/2}+24+276q^{1/2}+2048q+11202q^{3/2}+\cdots.$$ If we set $\tau=\sqrt{-163}$ and $q=e^{2\pi i\tau}=e^{-2\pi\sqrt{163}}$, then we have $$\mathfrak f(\sqrt{-163})^{24}=e^{\pi\sqrt{163}}+24+\frac{276}{e^{\pi\sqrt{163}}}+\frac{2048}{e^{2\pi\sqrt{163}}}+\frac{11202}{e^{3\pi\sqrt{163}}}+\cdots.$$ This gives the approximation $$e^{\pi\sqrt{163}}\approx\mathfrak f(\sqrt{-163})^{24}-24.$$ The improved accuracy of this approximation is due to the fact that the coefficient $276$ is smaller than the coefficient $196884$.

The only remaining mystery is why $\mathfrak f(\sqrt{-163})$ is the real root of the cubic polynomial $x^3-6x^2+4x-2$. One reference for this is the paper "On the Singular Values of Weber Modular Functions" by Yui and Zagier. From section 6 of the paper, we learn that there is a cubic polynomial \begin{align*} x^3-2\lambda x^2+2\mu x-2&=\left(x-\mathfrak f(\sqrt{-163})\right)\left(x-\frac{\sqrt2}{\mathfrak f\left(\frac{15+\sqrt{-163}}2\right)}\right)\left(x-\frac{\sqrt2}{\mathfrak f\left(\frac{-15+\sqrt{-163}}2\right)}\right)\\ &\approx(x-5.3186)(x-(0.3407+0.5099i))(x-(0.3407-0.5099i)) \end{align*} with $\lambda,\mu\in\mathbb{Z}$ (this uses the fact that $\mathbb{Q}(\sqrt{-163})$ has class number one). Then expanding the right hand side shows that $\lambda=3$ and $\mu=2$ as desired.

2

When dealing with exotic connections, sometimes it is helpful to take inspiration from Ramanujan. Thus, we'll bring in q-continued fractions, pi formulas, nested radicals, and sequences.


I. q-Continued fractions

Let $q = -e^{-\pi\sqrt{11}}.\,$ Then,

$$\frac{1}{T}+1 = \cfrac{(-q)^{-1/24}} {1 + \cfrac{q} {1-q + \cfrac{q^3-q^2} {1 + \cfrac{q^5-q^3} {1 + \cfrac{q^7-q^4} {1+\ddots}}}}}$$

where $T$ is the tribonacci constant. This is reminiscent of Ramanujan's cfracs which involve the golden ratio. And not surprisingly, the number $24$ appears again.

In general, for $q = -e^{-\pi\sqrt{n}}$ and positive integer $n$, then the cfrac is a radical usually of high degree for large $n$. However, for $n = (11,19,43,67,163),$

$$x_n=\cfrac{(-q)^{-1/24}} {1 + \cfrac{q} {1-q + \cfrac{q^3-q^2} {1 + \cfrac{q^5-q^3} {1 + \cfrac{q^7-q^4} {1+\ddots}}}}}$$

then $x_n$ is just the real root of the five cubics mentioned by Pegg, namely $x^3-2x^2+2x-2 = 0$, and the others.


II. Pi formulas

Given $v = x_{11} = \frac{1}{T}+1,$ where $T$ is still the tribonacci constant. Then,

$$\frac{1}{\pi} = \frac12\sum_{k=0}^\infty \frac{(2k)!^3}{k!^6}\frac{(2v-1)(4v-3)k+(2v-1)(v-1)-\frac12}{v^{24k}}$$

Is that our friend the number $24$ again? In fact, similar pi formulas can be found using $x_n$ for $n = (11,19,43,67,163).$


III. Nested radicals

We have the well-known,

$$\phi =\sqrt{1+\sqrt{1+\sqrt{1+\sqrt{1+\dots}}}}$$

with golden ratio $\phi$. The tribonacci version is,

$$\frac1{T-1}=\sqrt[3]{\frac12+\sqrt[3]{\frac12+\sqrt[3]{\frac12+\sqrt[3]{\frac12+\dots}}}}$$

Again, one can do the same for all cubic roots, but they may not be so pretty.


IV. Sequences

To avoid redundancy, more details can be found in this post.