1
$\begingroup$

A regular $ n$-gon is inscribed in the unit circle centered in $0$. We want to build an "almost" geodesic dome upon it this way: on each side of the $n$-gon we build an equilateral triangle whose dihedral angle with the $n$-gon is such that the distance from $0$ of the upper vertex's projection on the unit circle plane is $(1-1/n)$; over this "first floor" we connect all the upper vertices and we have a new smaller $n$-gon inscribed in the circle with radius $(1-1/n)$, so we repeat the process and this time the distance from $0$ of the upper vertices' projection on the unit circle plane is $(1-2/n)$ and so on. If the limit $n \rightarrow \infty$, which function fits the border of the dome? What is the maximum height?

I tried to solve this way: $n$ fixed and $r < n$, each upper vertex coordinate is

$$ \left( \frac {r+1}{n}, \sum_{m=0}^r \sqrt{ \left( \sqrt{3}\left(1-\frac mn\right)^2 \sin\left(\frac\pi n\right) \right)^2-\frac1{n^2}} \, \right)$$

With $n = 100 $ this is the plot made with Mathematica, but I wasn't able to find that function

enter image description here

$\endgroup$
6
  • 1
    $\begingroup$ Under the square root, do you have $1-\dfrac m{n}$ to the second or fourth power? $\endgroup$ Commented Jun 11 at 15:22
  • $\begingroup$ @IosifPinelis to the fourth power, because it's the hypothenuse squared. Thank you for your answer, I'm reading it. Do you think the expression I found is right? $\endgroup$
    – user967210
    Commented Jun 11 at 15:58
  • $\begingroup$ I will have to correct the answer then, when I have a chance. $\endgroup$ Commented Jun 11 at 17:01
  • $\begingroup$ So at a typical vertex, six triangles meet, but at the one vertex at the center, $n$ triangles meet. $\endgroup$ Commented Jun 11 at 17:07
  • 2
    $\begingroup$ I think I have now understood your derivation. In particular, it appears that $\frac{r+1}n$ is the distance of the projection of the new vertex after round $r=0,1,\dots$ from the circle. However, of course you should have $1-m/n$ instead of $(1-m/n)^2$. To quickly see why this is true, use a circle of radius $R$ (say centimeters) rather than $1$, and then check the physical dimensions of the terms. $\endgroup$ Commented Jun 12 at 2:10

1 Answer 1

3
$\begingroup$

The expression $$f_n(x):=\sqrt{\Big(\sqrt{3}\Big(1-\dfrac x{n}\Big) \sin\dfrac \pi{n} \Big)^2-\dfrac1{n^2}}$$ is defined for real $x\ge0$ iff $$x\le x_n:=n\Big(1-\frac1{a_n\sqrt3}\Big),$$ where $a_n:=n\sin\frac\pi n\to\pi$ (as $n\to\infty$). Also, the function $f_n$ is decreasing on the interval $[0,x_n]$. So, $0\le f_n\le f_n(0)=O(1/n)$. So, the sum, $\sum_{m=0}^r f_n(m)$ (corrected according to my comment), in your displayed expression differs from the integral $$I(n,r):=\int_0^r f_n(x)\,dx$$ for integers $r\in[0,x_n]$ only by $O(1/n)$.

Using the substitution $v=(1-x/n)a_n\sqrt3$, we get $$I(n,r)=\frac1{a_n\sqrt3}\int_{(1-r/n)a_n\sqrt3}^{a_n\sqrt3}\sqrt{v^2-1}\,dv \\ =\frac{G(a_n\sqrt3)-G((1-r/n)a_n\sqrt3)}{a_n\sqrt3},$$ where $$G(v):=\frac{1}{2} v \sqrt{v^2-1}-\ln \left(\sqrt{v-1}+\sqrt{v+1}\right),$$ an anti-derivative of $\sqrt{v^2-1}$.

It follows that, for each $$t\in(0,T),\quad \text{where }T:=1-\frac1{\pi\sqrt3},$$ the vertex $(\frac{r+1}n,\sum_{m=0}^r f_n(m))$ tends to $$(t,F(t))$$ as $n\to\infty$ and $r/n\to t$, where $$F(t):=\frac{G(\pi\sqrt3)-G((1-t)\pi\sqrt3)}{\pi\sqrt3}.$$

Here is the graph $\{(t,F(t))\colon0<t<T\}$:

enter image description here

$\endgroup$
6
  • $\begingroup$ Thanks for your detailed answer, I need to say that I found $ T:=1-\frac1{\pi\sqrt[4]{3}} $, so the profile is different. Am I wrong? $\endgroup$
    – user967210
    Commented Jun 11 at 16:17
  • 1
    $\begingroup$ @user967210 : Yes, your expression is incorrect -- my previous comment. $\endgroup$ Commented Jun 12 at 2:05
  • $\begingroup$ You are completely right! I don't know how that exponent came out! Thank you $\endgroup$
    – user967210
    Commented Jun 12 at 3:57
  • 1
    $\begingroup$ I have simplified the derivation (and also the expression) of $F(t)$. Please let me know if you have further questions or concerns. $\endgroup$ Commented Jun 12 at 12:18
  • 1
    $\begingroup$ @user967210 : I think so. The length of the limit curve will be $\int_a^b\sqrt{1+F'(t)^2}\,dt$ for some $a$ and $b$. So, you will have to integrate something like $\sqrt{A+Bv^2}$ in $v$, which is expressed in elementary functions. $\endgroup$ Commented Jun 14 at 13:11

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