14
$\begingroup$

I was experimenting with hypergeometric-like series and discovered the following conjecture (so far confirmed by more than $5000$ decimal digits): $$\sum_{n=0}^\infty\frac{\Gamma\!\left(n+\tfrac14\right)}{2^n\,(4n+1)^2\,n!}\stackrel{\color{gray}?}=\\ \frac{\Gamma\!\left(\tfrac14\right)\sqrt[4]2}{192}\left[\vphantom{\huge|}6\sqrt{2}\left(2\pi\ln2-\ln^22-8\operatorname{Li}_2\left(\tfrac1{\sqrt2}\right)\right)+3\psi^{(1)}\!\left(\tfrac18\right)-48G+\left(\vphantom{\large|}7\sqrt2-6\right)\pi^2\right],\tag{$\diamond$}$$ where $G$ is the Catalan constant, $\operatorname{Li}_2(x)$ is the dilogarithm and $\psi^{(1)}(x)$ is the trigamma function. Could you suggest any ideas how to prove it?


To see what approach I use to find conjectures like this, see my another question.


Update: I've found a generalization of this conjecture. See the corresponding Mathematica expression here. Hopefully, it can be simplified.

$\endgroup$
  • 1
    $\begingroup$ What was the origin of the conjecture? $\endgroup$ – Mark Viola Dec 21 '15 at 20:46
  • $\begingroup$ You could find a related question here. $\endgroup$ – user153012 Dec 21 '15 at 21:10
  • $\begingroup$ @Dr.MV I've described my approach to discovering conjectures in that question. $\endgroup$ – Vladimir Reshetnikov Dec 21 '15 at 21:11
11
+100
$\begingroup$

First note that $$\sum_{n=0}^\infty\frac{\Gamma\!\left(n+\tfrac14\right)}{2^n\,(4n+1)^2\,n!}=\Gamma\left(\frac14\right){}_3F_2\biggl[ \begin{array}{c}\frac14,\frac14,\frac14 \\ \frac54,\frac54\end{array};\frac12\biggr]$$

We will obtain an "elementary" expression for ${}_3F_2\biggl[ \begin{array}{c}\frac14,\frac14,\frac14 \\ \frac54,\frac54\end{array};z\biggr]$ with arbitrary $z$ using as a main tool the differentiation formula \begin{align}\left(z\frac{d}{dz}+\beta_k-1\right){}_pF_q\biggl[ \begin{array}{c}\alpha_1,\ldots,\alpha_p \\ \beta_1,\ldots,\beta_k,\ldots,\beta_q\end{array};z\biggr]&=\\ =\left(\beta_k-1\right) {}_pF_q\biggl[ \begin{array}{c}\alpha_1,\ldots,\alpha_p \\ \beta_1,\ldots,\beta_k-1,\ldots,\beta_q\end{array};z\biggr]&. \tag{$\spadesuit$}\end{align}

Lemma 1. We have $${}_2F_1\biggl[ \begin{array}{c}\frac14,\frac14 \\ \frac54\end{array};z\biggr]= \frac{z^{-\frac14}}4\left[e^{\frac{\pi i}4}\ln\frac{1+e^{-\frac{\pi i}4}t(z)}{1-e^{-\frac{\pi i}4}t(z)}+e^{-\frac{\pi i}4}\ln\frac{1+e^{\frac{\pi i}4}t(z)}{1-e^{-\frac{\pi i}4}t(z)}\right], \tag{$\clubsuit$}$$ where $t(z)=\left(\frac{z}{1-z}\right)^{\frac14}$.

Proof. Setting in ($\spadesuit$) $p=2$, $q=1$, $\alpha_1=\alpha_2=\frac14$, $\beta_1=\frac54$, we get $$\left(z\frac{d}{dz}+\frac14\right){}_2F_1\biggl[ \begin{array}{c}\frac14,\frac14 \\ \frac54\end{array};z\biggr]=\frac14{}_2F_1\biggl[ \begin{array}{c}\frac14,\frac14 \\ \frac14\end{array};z\biggr]=\frac{\left(1-z\right)^{-\frac14}}{4}.$$ The resulting 1st order ODE $zy'+\frac y4=\frac{\left(1-z\right)^{-\frac14}}{4}$ can be integrated by variation of integration constant: setting $y(z)=C(z)z^{-\frac14}$, one obtains $$C'(z)=\frac{z^{-\frac34}\left(1-z\right)^{-\frac14}}{4}\qquad \Longrightarrow\quad C(z)=\frac14\int z^{-\frac34}\left(1-z\right)^{-\frac14}dz.$$ The antiderivative can be computed in terms of elementary functions: setting $$z=\frac{t^4}{1+t^4},\qquad 1-z=\frac{1}{1+t^4},\qquad dz=\frac{4t^3dt}{(1+t^4)^2},\qquad t=\left(\frac{z}{1-z}\right)^{\frac14},$$ we get $$C(z)=\int\frac{dt}{1+t^4}=\frac14\left[e^{\frac{\pi i}4}\ln\frac{1+e^{-\frac{\pi i}4}t}{1-e^{-\frac{\pi i}4}t}+e^{-\frac{\pi i}4}\ln\frac{1+e^{\frac{\pi i}4}t}{1-e^{-\frac{\pi i}4}t}\right]+\operatorname{const}.$$ Fixing the integration constant via the condition ${}_2F_1\biggl[ \begin{array}{c}\frac14,\frac14 \\ \frac54\end{array};0\biggr]=1$ (in fact it suffices to know that ${}_2F_1$ is regular as $z\to 0$), we arrive at the representation ($\clubsuit$). In Prudnikov et al there is a formula (7.3.2.66, Vol. III) from which one should in principle be able to derive the same result after transformation of parameters. $\square$

Lemma 2. We have \begin{align}{}_3F_2\biggl[ \begin{array}{c}\frac14,\frac14,\frac14 \\ \frac54,\frac54\end{array};z\biggr]&=\frac{z^{-\frac14}}{4}\Biggl\{\int_0^{t(z)}\frac{\ln\left(1+t^{-4}\right)dt}{1+t^4} +\Biggr.\tag{$\heartsuit$} \\ &+\Biggl.\frac{\ln z}{4}\left[e^{\frac{\pi i}4}\ln\frac{1+e^{-\frac{\pi i}4}t(z)}{1-e^{-\frac{\pi i}4}t(z)}+e^{-\frac{\pi i}4}\ln\frac{1+e^{\frac{\pi i}4}t(z)}{1-e^{-\frac{\pi i}4}t(z)}\right]\Biggr\}, \end{align} where $t(z)=\left(\frac{z}{1-z}\right)^{\frac14}$.

Proof. The same procedure: setting in ($\spadesuit$) $p=3$, $q=2$, $\alpha_1=\alpha_2=\alpha_3=\frac14$, $\beta_1=\beta_2=\frac54$, we get $$\left(z\frac{d}{dz}+\frac14\right){}_3F_2\biggl[ \begin{array}{c}\frac14,\frac14 ,\frac14\\ \frac54,\frac54\end{array};z\biggr]=\frac14{}_3F_2\biggl[ \begin{array}{c}\frac14,\frac14,\frac14 \\ \frac54,\frac14\end{array};z\biggr]=\frac14{}_2F_1\biggl[ \begin{array}{c}\frac14,\frac14\\ \frac54\end{array};z\biggr],$$ which in turn implies that $${}_3F_2\biggl[ \begin{array}{c}\frac14,\frac14 ,\frac14\\ \frac54,\frac54\end{array};z\biggr]=\frac{z^{-\frac14}}{4}\int_0^{z}s^{-\frac34}{}_2F_1\biggl[ \begin{array}{c}\frac14,\frac14\\ \frac54\end{array};s\biggr]\,ds.$$ Substituting into this expression the representation found in Lemma 1 and making the change of variables $s\to t(s)=\left(\frac{s}{1-s}\right)^{\frac14}$, we arrive at $${}_3F_2\biggl[ \begin{array}{c}\frac14,\frac14 ,\frac14\\ \frac54,\frac54\end{array};z\biggr]=\frac{z^{-\frac14}}{4}\int_0^{t(z)}\left[e^{\frac{\pi i}4}\ln\frac{1+e^{-\frac{\pi i}4}t}{1-e^{-\frac{\pi i}4}t}+e^{-\frac{\pi i}4}\ln\frac{1+e^{\frac{\pi i}4}t}{1-e^{-\frac{\pi i}4}t}\right]\frac{dt}{t\left(1+t^4\right)}.$$ This clearly can be integrated in terms of dilogarithms. It is convenient to integrate first by parts using that the derivative of the expression in the square brackets is $\frac4{1+t^4}$, which yields ($\heartsuit$). $\square$

Corollary. Hypergeometric function ${}_3F_2\biggl[ \begin{array}{c}\frac14,\frac14,\frac14 \\ \frac54,\frac54\end{array};z\biggr]$ has an explicit expression in terms of elementary functions and dilogarithms, which should give the original statement as a particular case corresponding to $z=\frac12$.

Proof. It suffices to compute in explicit form the integral from the first line of ($\heartsuit$). Since the anti-derivative (check e.g. WolframAlpha) is given by a rather long expression, it is not written here. The resulting formula, however, should simplify to linked expression from the update of @VladimirReshetnikov. $\square$

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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