10

C. H. Brown found some rapidly converging infinite series for particular values of the gamma function , $$\frac {\Gamma\left(\tfrac 13\right)^6}{12\pi^4}=\frac 1{\sqrt{10}}\sum\limits_{k=0}^{\infty}\frac {(6k)!}{k!^3(3k)!}\frac {(-1)^k}{(3\cdot160^3)^k}$$$$\frac {\Gamma\left(\tfrac 14\right)^4}{128\pi^3}=\frac 1{\sqrt u}\sum\limits_{k=0}^{\infty}\frac {(6k)!}{k!^3(3k)!}\frac {1}{\big(2\sqrt2\,u^3v^6\big)^k}$$

where $u=273+180\sqrt2,$ and $v=1+\sqrt2$.

Questions:

  1. How did C. Brown derive these respective infinite formulas?
  2. Can you derive similar formulas for different gamma functions?

I couldn't help but notice that these formulas share a similarity to the Chudnovsky Algorithm$$\frac 1\pi=12\sum\limits_{k=0}^{\infty}\frac {(6k)!}{k!^3(3k)!}\frac {545140134k+13591409}{640320^{k+1/2}}$$So perhaps Brown used the J-function and modular forms to derive the values?$$j(\tau)=\frac 1q+744+196884q+21493760q^2+\cdots$$

Crescendo
  • 4,417
  • Looks somewhat like a BBP-style formula, the kind Ramanujan would solve in his sleep using Hypergeometric functions. I'm willing to bet the sums were discovered by a computer, and educated guess as to what such a sum might look like, and a lot of case checking until the computer found a likely match that was then formally proved – Brevan Ellefsen Jul 06 '17 at 22:25
  • Reference to the Brown results please. – Somos Jul 06 '17 at 22:48
  • 1
    @Somos Oh lol I just found it on Wikipedia. But all Wikipedia does is present the result, but not a proof. https://en.m.wikipedia.org/wiki/Particular_values_of_the_Gamma_function – Crescendo Jul 06 '17 at 22:52
  • @BrevanEllefsen I believe Brown may have been experimenting with the j-function. I’m curious though, since there are infinitely many $j(\tau)$, that he focused only on these two. The others are potentially interesting as well. – Tito Piezas III Oct 14 '23 at 18:53

3 Answers3

8

Back in 2011, I empirically found the general family for this type of formulas. The OP was right when he compared them to the Chudnovsky algorithm which uses the j-function $j(\tau)$.


I. j-function

The first has $j\big(\frac{1+\sqrt{-27}}{2}\big) = -3\times160^3.$

The second has $j(\sqrt{-16}) = 2\sqrt2\,(273+180\sqrt2)^3(1+\sqrt2)^6.$

A third should be $j(\sqrt{-6}) = 12^3\,(5+2\sqrt2)^3(1+\sqrt2)^2.$


II. Watson triple integrals

Brown's formulas can be connected to the "Watson triple integrals". (He missed the third one though.) Let,

\begin{align} I_1 &= \frac{1}{\pi^3} \int_{0}^{\pi} \int_{0}^{\pi} \int_{0}^{\pi} \frac{dx \, dy \, dz}{1-\cos x \cos y \cos z}\\ I_2 &= \frac{1}{\pi^3} \int_{0}^{\pi} \int_{0}^{\pi} \int_{0}^{\pi} \frac{dx \, dy \, dz}{3-\cos x \cos y -\cos x \cos z -\cos y \cos z}\\ I_3 &= \frac{1}{\pi^3} \int_{0}^{\pi} \int_{0}^{\pi} \int_{0}^{\pi} \frac{dx \, dy \, dz}{3-\cos x -\cos y -\cos z} \end{align}

then,

\begin{align} \frac{4^{4/3}}{3^2}\color{red}{I_2} &= \frac 1{\sqrt{10}}\sum\limits_{k=0}^{\infty}\frac {(6k)!}{k!^3(3k)!}\frac {(-1)^k}{(3\cdot160^3)^k} = \frac {\Gamma\left(\tfrac 13\right)^6}{12\pi^4}\\ \frac{1}{2^5}\color{red}{I_1} &= \frac 1{\sqrt u}\sum\limits_{k=0}^{\infty}\frac {(6k)!}{k!^3(3k)!}\frac {1}{\big(2\sqrt2\,u^3\,(1+\sqrt2)^6\big)^k} = \frac {\Gamma\left(\tfrac 14\right)^4}{128\pi^3}\\ \frac{1}{2^{1/2}}I_3 &= \frac 1{\sqrt w}\sum\limits_{k=0}^{\infty}\frac {(6k)!}{k!^3(3k)!}\frac {1}{\big(12^3w^3\,(1+\sqrt2)^2\big)^k} = \frac {\Gamma\left(\tfrac 1{24}\right)\Gamma\left(\tfrac 5{24}\right)\Gamma\left(\tfrac 7{24}\right)\Gamma\left(\tfrac {11}{24}\right)}{32\sqrt3\,\pi^3} \end{align}

where $u=273+180\sqrt2,$ and $w=5+2\sqrt2$.


III. Gamma functions

For the first, given $\tau = \big(\frac{1+\sqrt{-d}}{2}\big)$ where $d>3$, a general formula is the simple sum-product identity,

$$\frac1{\big({-j(\tau)}\big)^{1/6}}\sum_{n=0}^\infty \frac{(6n)!}{(3n)!(n!^3)} \frac1{\big(j(\tau)\big)^n} = \frac1{(2\pi)d}\prod_{m=1}^{d-1} \Big[\Gamma\big(\tfrac m d\big)\Big]^{\big(\frac{-d}{m}\big)}$$

where the gamma functions's exponent $\big(\frac{-d}{m}\big)$ is the Kronecker symbol. For example, let $d = 7$, so $j\big(\frac{1+\sqrt{-7}}{2}\big)=-15^3$. Thus,

$$\frac1{\big({15^3}\big)^{1/6}}\sum_{n=0}^\infty \frac{(6n)!}{(3n)!(n!^3)} \frac1{\big({-15^3}\big)^n} = \frac1{7(2\pi)}\frac{\Gamma\big(\tfrac17\big)\Gamma\big(\tfrac27\big)\Gamma\big(\tfrac47\big)}{\Gamma\big(\tfrac37\big)\Gamma\big(\tfrac57\big)\Gamma\big(\tfrac67\big)} $$

and so on.

  • Related post on the Watson triple integrals. – Tito Piezas III Oct 14 '23 at 10:27
  • I find these results quite amazing and also I'm upset that I am the only one who up voted this while I got votes for just pointing to some links. – marty cohen Oct 14 '23 at 17:04
  • @martycohen Thanks, Marty. With the speed that posts move, the people who appreciate such posts probably didn’t see it. – Tito Piezas III Oct 14 '23 at 18:36
  • I saw it, thanks to the comment from Tito. Happy I saw it too :) really pretty results – Brevan Ellefsen Oct 14 '23 at 18:59
  • 1
    These formulas are the hypergeometric series for $(2K(k)/\pi)^2$ in terms of Ramanujan's class invariants $G_n, g_n$ namely $(2K/\pi)^2=(1-4G_n^{-24})^{-1/2}{}3F_2(1/6,5/6,1/2;1,1;-27G_n^{48}/(G_n^{24} -4)^3)$ and $(2K/\pi)^2=(k'^4+16k^2)^{-1/2}{}_3F_2(1/6,5/6,1/2;1,1;27g_n^{48}/(g_n^{24}+4)^3)$. The first series in question is for $G{27}$ and second series is for $g_{16}$. – Paramanand Singh Oct 18 '23 at 10:33
  • @ParamanandSingh Thanks for the info. In the webaerchive link, I also noticed the connection to $K(k)$ but used gamma functions above since the OP inquired about them. I used $\tau = \frac{1+\sqrt{-n}}{2}$, hence $G_n$, but your formula using $g_n$ fills in the other half. P.S. As you can tell, the third series (in my answer) is $g_6$. – Tito Piezas III Oct 18 '23 at 12:23
  • I checked your webarchive link in detail and I find it strongly related to Berndt's paper "Ramanujan's theory of elliptic functions to alternative bases". The earliest development was done by Borwein brothers who found series for $(2K/\pi)^2$ in terms of ${}3F{2}(r,1-r,1/2;1,1;c_r(k))$ for $r=1/2,1/3,1/4,1/6$ and suitable functions $c_r(k) $. Later Berndt found all this hidden in notebooks of Ramanujan and he wrote his paper based on that. In particular it should help to find an expression for $r_{3A}(\tau)$ given in your link. – Paramanand Singh Oct 19 '23 at 05:48
  • @ParamanandSingh Thanks! Yes, i was having trouble with the case $r = 1/3$. Maybe I should revisit these formulas. – Tito Piezas III Oct 19 '23 at 06:02
  • If you have access to Browein's book "Pi and the AGM", it contains the hypergeometric series for $(2K/\pi)^2$. The formula is complicated for $r=1/3$ and you may details in the book from page 177-191 – Paramanand Singh Oct 19 '23 at 06:13
5

That Wikipedia link points to http://www.iamned.com/math/ which has a large number of amazing formulae.

That, in turn, points to http://iamned.com/math/infiniteseries.pdf titled "An Algorithm for the Derivation of Rapidly Converging Infinite Series for Universal Mathematical Constants".

marty cohen
  • 110,450
3

We have derived some of this values you are searching in this question: Method to obtain values of Gauss hypergeometric function.

That allows to get: $$\sum_{n=0}^{\infty}\frac{(3n)!(2n)!(-1)^n}{n!^5}\left(\frac{1}{8640} \right)^n=\frac{9\cdot 5^{2/3}\cdot\Gamma(1/3)^6}{100\pi^4},\tag{1}$$ $$\sum_{n=0}^{\infty}\frac{(4n)!(-1)^n}{n!^4}\left(\frac{1}{6635520} \right)^n=\frac{3\cdot 5^{1/4}\cdot\Gamma(1/4)^4}{25 \pi^3}.\tag{2}$$ Using Clausen's formula: $$\left(\sum_{n=0}^{\infty}\frac{f(a)f(b)}{f(2b)n!}x^n\right)^2=(1-x)^{-a}\sum_{n=0}^{\infty}\frac{f(a)f(b)f(2b-a)}{f(b+1/2)f(2b)n!}\left(\frac{x^2}{4(x-1)} \right)^n\tag{3}$$ Where $f(x)=\frac{\Gamma{(x+n)}}{\Gamma{(x)}}=(x)_{n},$ is the Pochhammer symbol. Some values are obtained from Ebisu's work. There are formulas connecting the four known theories in which binomial coefficients are of the form: $$\frac{(2n)!^3}{n!^6}\hspace{.5cm}\frac{(3n)!(2n)!}{n!^5}\hspace{.5cm}\frac{(4n)!}{n!^4}\hspace{.5cm}\frac{(6n)!}{(3n)!n!^3} \tag{4}$$ Perhaps one of the most simple is: $$\sum_{n=0}^{\infty}\frac{(-1)^n(2n)!^3}{2^{9n}n!^6}=\frac{\Gamma{(1/4)}^4}{4\sqrt{2}\pi^3}\tag{5}$$ which you can find here: Integrals of elliptic integrals. We also arrived to: $$\sum_{n=0}^{\infty}\frac{(-1)^n(2n)!^3(2n+1)}{2^{9n}n!^6}=\frac{2\sqrt{2}}{3\pi}+\frac{\Gamma(\frac{1}{4})^4}{6\sqrt{2}\pi^3}\tag{6}$$ and $$\sum_{n=0}^{\infty}\frac{(-1)^n(6n+1)(2n)!^3}{2^{9n}n!^6}=\frac{2\sqrt{2}}{\pi}.\tag{7}$$ Another example is obtained here also here On Ramanujan's fastest series. in an elementary way: $$\sum_{n=0}^{\infty}\frac{(4n)!}{648^n n!^4}=\frac{3\Gamma{(\frac{1}{4})^4}}{16\pi^3}\tag{8}$$ and the associated series for $1/\pi$: $$\sum_{n=0}^{\infty}\frac{(7n+1)(4n)!}{648^n n!^4}=\frac{9}{2\pi}.\tag{9}$$

A very interesting paper by Zudilin https://arxiv.org/abs/1210.0269 (Lost in translation) gives the links between the four theories. So $(5)$ and $(8)$ and $(7)$ and $(9)$ are the same in the sense of Zudilin's paper.

Finally Piezas's answer is in fact Chowla-Selberg formula with gives the explicit evaluation of the complete elliptic of first kind in terms of pi and gamma functions.

User-Refolio
  • 1,088
  • 4
  • 29
  • 1
    These formulas are based on Ramanujan's work which uses a hypergeometric series for $(2K/\pi)^2$ and the identity $P(q^2)=(2K/\pi)^2(1-2k^2)+\frac{3kk'^2}{2}\frac{d}{dk}(2K/\pi)^2$. Then we compare it with $P(q^2)=-\frac{3}{\log q} +A_n(2K/\pi)^2$ where $q=\exp(-\pi\sqrt{n}),n\in\mathbb{Q}^+$ and $A_n$ is some suitable algebraic number dependent on $n$. – Paramanand Singh Mar 26 '25 at 17:12
  • 1
    BTW I wrote some long answers for my own question on Chudnovsky series on MO. See https://mathoverflow.net/questions/384027/a-formula-in-ramanujans-lost-notebook-and-its-connection-with-chudnovsky-series if you get some free time. – Paramanand Singh Mar 26 '25 at 17:16
  • Amazing work @ParamanandSingh for first I wanted to list some of your work in my answer that I can improve with full details using the most elementary approach (as we have seen that in recent answers of mine) and then extend it with elliptic functions, modular forms andWZmethod. But this is a lot of work. Maybe you can give an answer wich complements my answer that will cover: $k_{1}$, $k_{2}$, $k_{3}$ and $k_{4}$ wich were known since Legendre and with the application of Clausen's formula for $K^2(k)$ and $E(k)K(k)$ gives the easiest series for $1/\pi$ without using Jacobi and Ramanujan ideas – User-Refolio Mar 27 '25 at 08:35
  • Finally @ParamanandSingh what's the difference between MO and MSE? Seems that in MO there are more seniority? Best. – User-Refolio Mar 27 '25 at 08:37
  • Your approach involves a direct evaluation of $E(k_n) $ which is possible using value of $E(k_1)=E(1/\sqrt{2})$ and Landen transform when $n$ is power of $2$. But I am not sure of we can get the value of $E(k_2),E(k_3)$ without using Ramanujan's ideas (at least I am not aware of it). – Paramanand Singh Mar 27 '25 at 14:30
  • @ParamanandSingh We can establish the algeabric numbers that relate $E(k_{2})$ and $E(k_{3})$ with $K(k_{2})$ and $K(k_{3})$ respectively using only integrals without knowing the explicit values of $K(k_{2})$ and $K(k_{3})$. I will post the details. – User-Refolio Mar 28 '25 at 08:54
  • An example of this is here: https://math.stackexchange.com/questions/3107321/how-to-evaluate-sum-n-0-infty-frac-1n4n12n326nn6/4898944#4898944 As you see for $k_{2}$ Landen's transformation works fine (modular equation of degree 2 in the language of elliptic functions) this gives the link of $K(k_{2})$ and $E(k_{2})$ without knowing explicit values when we translate to series. – User-Refolio Mar 28 '25 at 10:01
  • The link for $K(k_{3})$ and $E(k_{3})$ is harder but with integral manipulations we can link $E(k_{3})$ and $K(k_{3})$ without knowing explicti value of $K(k_{3})$ this is the part I have not post but I know to get. As I said I will improve this answer. Best. – User-Refolio Mar 28 '25 at 10:04
  • @ParamanandSingh Did you have time to check the solution proposed to Bauer's series? As you see Landen transformation works with $k_{2}=\sqrt{2}-1$ combined with the series representation for $K^2(k)$ and $E(k)K(k)$ and $E(k')K(k')$ this approach works only for $k_{2}$. – User-Refolio Apr 05 '25 at 10:28