41

In JIMS 4, p.78, Question 359 was asked by Ramanujan. (See The Problems Submitted by Ramanujan to the Journal of the Indian Mathematical Society, p. 9, by Bruce Berndt, et al.) If,

$$\sin(x+y) = 2\sin\big(\tfrac{1}{2}(x-y)\big)\tag1$$

$$\sin(y+z) = 2\sin\big(\tfrac{1}{2}(y-z)\big)\tag2$$

prove that,

$$\big(\tfrac{1}{2}\sin x\cos z\big)^{1/4}+\big(\tfrac{1}{2}\cos x\sin z\big)^{1/4} =\big(\sin 2y)^{1/12}\tag3$$

The example by Ramanujan was,

$$\begin{aligned} x &= \frac{\pi-\arcsin\big((\sqrt{5}-2)^3(4+\sqrt{15})^2\big)}{2}=1.094\dots\\ y &=\frac{\arcsin\big(\sqrt{5}-2\big)}{2}=0.119\dots\\ z &=\frac{\arcsin\big((\sqrt{5}-2)^3(4-\sqrt{15})^2\big)}{2}=0.0001\dots \end{aligned}$$

Ten years later, a 3-page proof was given in JIMS 15, p.114-117.


I got an email asking if there was a shorter proof. Considering the problem, I observed the following. Given the quartic,

$$a^3w^4+(1-3a^2)w^3+3a(1-a^2)w^2+a^2(3-a^2)w-a=0\tag4$$ with $a=\tan(\color{blue}y/4)$. Define, $$x=4\tan^{-1} u\\z=4\tan^{-1} v$$ where $u,v$ are two appropriate roots of the quartic, then we get the same bizarre relation as Ramanujan,

$$\big(\tfrac{1}{2}\sin x\cos z\big)^{1/4}+\big(\tfrac{1}{2}\cos x\sin z\big)^{1/4} =\big(\sin 2\color{blue}y)^{1/12}\tag5$$

Equivalently, those two roots $u,v$ obey,

$$\left(\frac{1}{2}\,\frac{4(u-u^3)}{(u^2+1)^2}\frac{v^4-6v^2+1}{(v^2+1)^2} \right)^{1/4}+\left(\frac{1}{2}\,\frac{4(v-v^3)}{(v^2+1)^2}\frac{u^4-6u^2+1}{(u^2+1)^2} \right)^{1/4}=\big(\sin 2\color{blue}y\big)^{1/12}\tag6$$

For example, let $y=1$, so $a=\tan(1/4)$, then $u,v$ are the two real roots of (4).

Question: Anyone knows a short proof for (3) and (6)?

  • 3
    I can't read the post, part of it is cut off. Is this something everyone's facing? – Ishan Banerjee Feb 21 '13 at 09:49
  • 15
    If Ramanujan couldn't solve it... – Nathaniel Bubis Feb 21 '13 at 10:04
  • 28
    @nbubis can you imagine if Ramanujan had lived long enough to be on math.se? – Alexander Gruber Feb 21 '13 at 10:48
  • 1
    @AlexanderGruber - Just saying :) maybe it should be on MO since it seems to involve research level mathematics. – Nathaniel Bubis Feb 21 '13 at 10:51
  • @IshanBanerjee, yes: I've the same problem. The post is cut off. – DonAntonio Feb 21 '13 at 13:44
  • 1
    If you use Google Chrome then the post is fine, but if you an old version of MS Explorer then it is cut-off. Don't know why. – Tito Piezas III Feb 27 '13 at 03:32
  • Let's say I am considering proving (3)........What I could see is that $x=75 degrees$, $y=15 degrees$ and $z=435 degrees$ satisfy equations (1) and (2)...But they don't satisfy (3)....Can anybody please check and see if there is actually something wrong with (3) or not? – seavoyage Jul 14 '16 at 18:20
  • @AkashBajaj: I've edited the post to add a link to Berndt's paper and the page where the proposed relation is given. Hm, you're right, I can't get your example to work nor Ramanujan's in the cited paper. It's been a long time, maybe I forgot to use an assumption. – Tito Piezas III Jul 14 '16 at 22:29
  • Thanks @TitoPiezasIII ....also, it would be great if you could post a link to that 3-page proof you mentioned above....if you have it.....that would help us check how they verified the solutions... – seavoyage Jul 15 '16 at 10:22
  • It's ok....I got that now...the angles must be acute – seavoyage Jul 15 '16 at 19:52
  • @AkashBajaj: I edited again to add Ramanujan's example. I find the angles of his $x,y,z$ (multiplied by $180/\pi$) as approximately $64.6,, 6.8,, 0.006$ so they are all acute. May I know the reference regarding how $x,y,z$ should have acute angles? – Tito Piezas III Jul 16 '16 at 01:08
  • Yeah...I got the same ones...I don't have a reference....I guessed that would be the assumption since my values didn't work out.....I am "hoping" that the angles must be acute... :\ – seavoyage Jul 16 '16 at 16:02
  • @AkashBajaj: Ah, I see :\ But it's good as your example can be used to refine the problem – Tito Piezas III Jul 16 '16 at 21:53

2 Answers2

11

Thanks to a deleted answer by G. Manco, we finally have a partial answer after four years to a closed-form solution $x,y,z$ to, $$\big(\tfrac{1}{2}\sin x\cos z\big)^{1/4}+\big(\tfrac{1}{2}\cos x\sin z\big)^{1/4} =\big(\sin 2y)^{1/12}\tag3$$ in terms of the Ramanujan G function. Using the Dedekind eta function $\eta(\tau)$, this can be calculated in Mathematica as, $$G_n=\frac{2^{-1/4}\,\eta^2(\sqrt{-n})}{\eta\big(\tfrac{\sqrt{-n}}{2}\big)\,\eta(2\sqrt{-n})}\tag4$$ However, because of the constraints $(1),(2)$ in the post, there is a difference when $1\leq n\leq9$ and $n\geq9$. Thus,

If $1\leq n\leq9$: $$\begin{aligned} x &=\pi/2-\tfrac12\arcsin G_{n/9}^{-12}\\ y &=\tfrac12\arcsin G_{n}^{-12}\\ z &=\tfrac12\arcsin G_{9n}^{-12} \end{aligned}$$ If $n\geq9$: $$\begin{aligned} x &=\tfrac12\arcsin G_{n/9}^{-12}\\ y &=\tfrac12\arcsin G_{n}^{-12}\\ z &=\tfrac12\arcsin G_{9n}^{-12} \end{aligned}$$ We recover Ramanujan's solution $x,y,z \approx 1.094,\; 0.119,\; 0.0001$ by plugging in $n=5$ to the first set of formulas. Also, since,

$$\sin(\tfrac12\arcsin \lambda)=\tfrac12\big(\sqrt{1+\lambda}-\sqrt{1-\lambda}\big)$$ $$\cos(\tfrac12\arcsin \lambda)=\tfrac12\big(\sqrt{1+\lambda}+\sqrt{1-\lambda}\big)$$ then $(3)$ can be expressed in radical form,

$$\left(\tfrac12\Big(\tfrac{\sqrt{1+\alpha}+\sqrt{1-\alpha}}2\Big)\Big(\tfrac{\sqrt{1+\beta}\,\color{blue}\pm\,\sqrt{1-\beta}}2\Big)\right)^{1/4}+\left(\tfrac12\Big(\tfrac{\sqrt{1+\alpha}-\sqrt{1-\alpha}}2\Big)\Big(\tfrac{\sqrt{1+\beta}\,\color{blue}\mp\,\sqrt{1-\beta}}2\Big)\right)^{1/4}=\gamma^{1/12}$$ where $\alpha = G_{n/9}^{-12},\;\beta= G_{9n}^{-12},\;\gamma= G_{n}^{-12}$ and the $+,-$ case for $n\leq9$ and $-,+$ case for $n\geq9$.

P.S. The second part of the question is still a bit tricky to prove.

10

An elegant solution to question 359 submitted by Ramanujan to Journal of the Indian Mathematical Society.

The three equations (1)-(3) are three modular equations: the first two are of third degree and the last one of ninth degree. These equations are in Ramanujan's notebooks: http://www.imsc.res.in/~rao/ramanujan/index.html.

If $\alpha$ be of the $1^{st}$, $\beta$ the $3^{rd}$ and $\gamma$ the $9^{th}$ degree then.

$\sqrt{\alpha(1-\beta)}$+$\sqrt{\beta(1-\alpha)}$=$2(\alpha\beta(1-\alpha)(1-\beta))^\frac {1} {8}$$\tag1$

$\sqrt{\gamma(1-\beta)}$+$\sqrt{\beta(1-\gamma)}$=$2(\beta \gamma (1-\beta)(1-\gamma))^\frac{1} {8}$$\tag2$

and

$(\alpha(1-\gamma))^\frac{1}{8}$+ $(\gamma(1-\alpha))^\frac{1}{8}$=$2^\frac{1}{ 3}$($\beta(1-\beta))^\frac{1}{24}$ $\tag3$

Here is my interpretation. The reading key is:

$\sqrt{\alpha}=\sin{x}$, $\sqrt{\beta}=\sin{y}$, and $\sqrt{\gamma}=\sin{z}$

moreover

$\sqrt{1-\alpha}=\cos x$, $\sqrt{1-\beta}=\cos y$ and $\sqrt{1-\gamma}=\cos z$.

and with class invariants

$4\alpha(1-\alpha)G_{\frac n 9}^{24}=1$, $4\beta(1-\beta)G_{n}^{24}=1$, and $4\gamma(1-\gamma)G_{9n}^{24}=1$

or

$\sin{2x}.G_{\frac {n} {9}}^{12}=1$ , $\sin{2y}.G_{n}^{12}=1$ , and $\sin{2z}.G_{9n}^{12}=1$.

For $n=5$ we have

$\sin 2x=2\sqrt{\alpha(1-\alpha)} =G_{\frac {5} {9}}^{-12}$=$((\sqrt 5+2)^{\frac 1 4}(4-\sqrt{15})^{\frac 1 6})^{-12}=(\sqrt 5-2)^3(4+\sqrt 15)^2$

$\sin 2y=2\sqrt{\beta(1-\beta)}=G_{5}^{-12}=\big((\frac {\sqrt 5+1}{2})^{\frac1 4}\big)^{-12}=(\sqrt 5-2)$

$\sin 2z=2\sqrt{\gamma(1-\gamma)}=G_{45}^{-12}=((\sqrt 5+2)^{\frac 1 4}(4+\sqrt{15})^{\frac 1 6})^{-12}=(\sqrt 5-2)^3(4-\sqrt 15)^2$

$(\frac 1 2 \sin x \cos z)^{\frac 1 4}=\frac{ 1} {\sqrt 2}(\sqrt 5-2)^{\frac 3 4}\big(\sqrt \frac {7+3\sqrt 5}{4}+\sqrt \frac {3+3\sqrt 5}{4}\big)$

$(\frac 1 2 \sin z \cos x)^{\frac 1 4}=\frac{ 1} {\sqrt 2}(\sqrt 5-2)^{\frac 3 4}\big(\sqrt \frac {7+3\sqrt 5}{4}-\sqrt \frac {3+3\sqrt 5}{4}\big)$

adding we have

$\frac{ 1} {\sqrt 2}(\sqrt 5-2)^{\frac 3 4}\big(2\sqrt \frac {7+3\sqrt 5}{4}\big)=(\sqrt 5-2)^{\frac 1 {12}}=(\frac {\sqrt 5-1}{2})^{\frac1 4}$

This result is also equal to

$(\sin 2y )^{\frac {1} {12}}=\frac {1} {G_{5}}=(\frac {\sqrt 5-1}{2})^{\frac{1} {4}}$

I calculated several solutions for $n=1,1/2,2,3,1/3,4,1/4,5,1/5,6,1/6,7,1/7,8,1/8,9,1/9,10,1/10,15,1/15.$

For $n=12$ we have

$\sin x$=$(\sqrt{2}+1)^2(\sqrt{3}-\sqrt{2})^2$

$\sin y$=$(\sqrt{2}-1)^2(\sqrt{3}-\sqrt{2})^2$

$\sin z$=$\frac{2-\sqrt{2+\sqrt{\delta}}} {2+\sqrt{2+\sqrt{\delta}}}$

with

$\delta=3-100.2^{\frac{1} {3}}+80.2^{\frac{2} {3}}$.