Reshetnikov gave the remarkable evaluation,

\begin{align} I&= \int_0^1\arctan{_4F_3}\left(\frac15,\frac25,\frac35,\frac45;\frac24,\frac34,\frac54;\frac{1}{64}\,x\right)\,dx \\ &=\frac{3125}{48}\left(5+3\pi+6\ln2-3\alpha^4+4\alpha^3+6\alpha^2-12\alpha\\-12\left(\alpha^5-\alpha^4+1\right)\arctan\frac1\alpha-6\ln\left(1+\alpha^2\right)\right)\\ &=0.7857194\dots \end{align} where $\alpha$ is a quartic root. However, it seems this can be simplified a bit and generalized.

I. $p=5$: Given,

$$I(n)=\int_0^1\arctan\,_4F_3\left(\frac15,\frac25,\frac35,\frac45;\,\frac24,\frac34,\frac54;\,\frac{n}{4^4}\,x\right)\,\mathrm dx$$ Is it true that, in general, $$\frac{12n}{5^5}\, I(n) = -12\Big(1\color{blue}-\frac{n}{5^5}\Big)\arctan\frac1\alpha+6\ln\Big(\frac{2}{1+\alpha^2}\Big)+3\pi+P(\alpha)$$ where $P(\alpha)=(1-\alpha)^3(5+3\alpha)$ and $\alpha$ is the largest positive root of the quintic, $$\alpha^5-\alpha^4+\frac{n}{5^5}=0$$ provided real number $n<4^4\,$? (Note: By sheer coincidence, the choice of $n=4$ in the other post caused the quintic to factor.)

II. $p=7$: Given,

$$J(n)=\int_0^1\arctan\,_6F_5\left(\frac17,\frac27,\frac37,\frac47,\frac57,\frac67;\,\frac26,\frac36,\frac46,\frac56,\frac76;\,\frac{n}{6^6}\,x\right)\,\mathrm dx$$ is it true that, $$\frac{60n}{7^7}\, J(n) = 60\Big(1\color{blue}+\frac{n}{7^7}\Big)\arctan\frac1\alpha-30\ln\Big(\frac{2}{1+\alpha^2}\Big)-15\pi-P(\beta)$$ where $P(\beta) = (1-\beta)^2(27-6\beta-9\beta^2+8\beta^3+10\beta^4)$ and $\beta$ is the largest positive root of, $$\beta^7-\beta^6+\frac{n}{7^7}=0$$ provided real $n<6^6\,$?


  1. How do we prove the two conjectured equalities?
  2. What's the formula for $p=3$? (Mathematica takes too long to evaluate the integral that I couldn't use an integer relations subroutine on it.)
  • $\begingroup$ But Reshetnikov didn't show a complete solution. Just the final answer. $\endgroup$ – Zaid Alyafeai Jan 3 '17 at 5:56
  • $\begingroup$ @ZaidAlyafeai: Yes, but it's a great final answer. He used an $\alpha$ that turns out to be the quartic factor of the quintic in this post. How he came across the correct root and the correct form is remarkable. I just found the $p=7$ by analogy. $\endgroup$ – Tito Piezas III Jan 3 '17 at 6:04
  • 2
    $\begingroup$ It looks like one can solve the general problem here for an arbitrary $k$. If $$I_k(n) = \int_{0}^{1}\arctan\left._k\right.F_{k-1}\left(\frac{1}{k+1},\frac{2}{k+1}\cdots,\cdots,\frac{k}{k+1},\frac{k+1}{k}; \frac{nx}{k^k}\right){\rm d}x$$ then we should have something like $$I_k(n) = \arctan\left(\frac{1}{z_0}\right) + \frac{(k+1)^{k+1}}{n}\int_{z_0}^1\frac{w^{k+1}-w^k}{1+w^2}{\rm d}w$$ where $z_0$ is a solution to $z_0^{k+1} - z_0^k + \frac{n}{(k+1)^{k+1}} = 0$. $\endgroup$ – Winther Jan 3 '17 at 22:38

I think you should start by the following general form

$${}_{k}F_{k-1}\left(\frac{1}{k+1} ,\cdots ,\frac{k}{k+1};\frac{2}{k} \cdots ,\frac{k-1}{k},\frac{k+1}{k};\left( \frac{m(1-m^k)}{f_k}\right)^k \right) = \frac{1}{1-m^k}$$


$$f_k \equiv \frac{k}{(1+k)^{1+1/k}}$$

Put $k=4$

$${}_{4}F_{3}\left(\frac{1}{5} ,\cdots ,\frac{4}{5};\frac{2}{4} \cdots ,\frac{3}{4},\frac{5}{4};\left( \frac{m(1-m^4)}{f_4}\right)^4 \right) = \frac{1}{1-m^4}$$

The argument simplifies to

$$ \left(\frac{m(1-m^4)}{\frac{4}{5^{1+1/4}}}\right)^4 = \frac{5^5 \,m^4(1-m^4)^4}{4^4}$$

Hence we have

$${}_{4}F_{3}\left(\frac{1}{5} ,\cdots ,\frac{4}{5};\frac{2}{4} \cdots ,\frac{3}{4},\frac{5}{4};\frac{5^5 (1-m)m^4}{4^4} \right) = \frac{1}{m}$$

Now suppose we want to find

$$\int \arctan\,_4F_3\left(\frac15,\frac25,\frac35,\frac45;\,\frac24,\frac34,\frac54;\,\frac{n}{4^4}\,x\right)\,\mathrm dx$$

Use $nx = 5^5m^4(1-m)$

$$\frac{5^5}{n}\int (4m^3(1-m)-m^ 4)\arctan\,_4F_3\left(\frac15,\frac25,\frac35,\frac45;\,\frac24,\frac34,\frac54;\,\frac{5^5 (1-m)m^4}{4^4}\right)\,\mathrm dm$$

This simplifies to

$$\frac{5^5}{n}\int (4m^3(1-m)-m^ 4)\arctan\,\frac{1}{m}\,\mathrm dm$$

The anti-derivative of this is elementary.

Note that when we use substittution we will need the roots of

$$m^5-m^4+\frac{n}{5^5}= 0$$

As conjectured by the OP.

Using the value $n=4$ we can verify Reshetnikov result

\begin{align}\int_0^1\arctan{_4F_3}\left(\frac15,\frac25,\frac35,\frac45;\frac24,\frac34,\frac54;\frac{1}{64}\,x\right)\,dx &= \frac{5^5}{4}\int^{\alpha}_1 (4x^3(1-x)-x^ 4)\arctan \left(\frac{1}{x} \right)\,\mathrm dx\\&=0.7857194\dots \end{align}

Where $\alpha$ is the largest positive root of

$$m^5-m^4+\frac{4}{5^5}= 0$$

  • $\begingroup$ Ok, but perhaps you should change your variable $x$ to the more generic $m$, as it will confuse/conflate the use of $F(x)dx$ in the post. $\endgroup$ – Tito Piezas III Jan 3 '17 at 6:51
  • $\begingroup$ @TitoPiezasIII, yes sure. $\endgroup$ – Zaid Alyafeai Jan 3 '17 at 6:56
  • 1
    $\begingroup$ Ah, beautiful! Mathematica can integrate $$\int^{\alpha}_1 (4x^3(1-x)-x^ 4)\arctan\left(\frac{1}{x} \right)\,\mathrm dx$$ for generic $\alpha$. I can now see why using $\ln(1+\alpha^2)$ and the other expressions make sense. $\endgroup$ – Tito Piezas III Jan 3 '17 at 12:29
  • $\begingroup$ do you have a proof for your first eqn.? $\endgroup$ – tired Jan 3 '17 at 13:24
  • $\begingroup$ @tired, I never worked with the generalized hyper geometric function. I will try to provide a proof for k=2. $\endgroup$ – Zaid Alyafeai Jan 3 '17 at 18:36

The result does generalize. We will derive a closed formula for the integral $$I_k(n) = \int_0^1\arctan\left[_{k}F_{k-1}\left(\frac{1}{k+1},\cdots, \frac{k}{k+1},\frac{2}{k},\cdots,\frac{k+1}{k}; \frac{nx}{k^k}\right)\right]\,{\rm d}x$$ valid for $|n| \leq k^k$.

For this particular combination of arguments the hypergeometrical function simplifies greatly, see property $(25)$ here$^{(1)}$. Using this we can perform a substitution $\frac{nx}{(k+1)^{k+1}} = y^k(1-y^k)^k$ and the integrand simplifies to $\frac{(k+1)^{k+1}}{n}\frac{1}{1-y^k}\frac{d}{dy}\left(y^k(1-y^k)^k\right)$. Now performing integration by parts and the substitutions $y^k\to y \to 1-y$ we arrive at

$$I_k(n) = \frac{\pi}{2} - \arctan(z_0) + \frac{(k+1)^{k+1}}{n}\int_{z_0}^1\frac{w^{k+1}-w^k}{1+w^2}{\rm d}w$$

where $z_0$ is the largest positive root of $z_0^{k+1} - z_0^k + \frac{n}{(k+1)^{k+1}} = 0$. The integral above can be evaluated in closed form using

$$\matrix{\int \frac{w^{2k}}{1+w^2}{\rm d}w &=& (-1)^k\left[\arctan(w) + F_k(w)\right] + C\\\int \frac{w^{2k+1}}{1+w^2}{\rm d}w &=& (-1)^k\left[\frac{1}{2}\log(1+w^2) + G_k(w)\right] + C}$$ where $F_k(x) = \sum_{i=1}^{k}\frac{(-1)^i x^{2i-1}}{2i-1}$ and $G_k(x) = \sum_{i=1}^{k}\frac{(-1)^i x^{2i}}{2i}$. These relations are easily proven by differentation and summing a geometrical series. This leads to

$$I_{2k}(n) = \frac{\pi}{2} - \arctan(z_0) + \frac{(-1)^k(2k+1)^{2k+1}}{n}\left[\frac{1}{2}\log\left(\frac{2}{1+z_0^2}\right) - \frac{\pi}{4} + \arctan(z_0)\right] + \frac{(-1)^k(2k+1)^{2k+1}}{n}\left[G_k(1) - G_k(z_0) - F_k(1) + F_k(z_0)\right]$$ $$I_{2k+1}(n) = \frac{\pi}{2} - \arctan(z_0) + \frac{(-1)^{k}(2k+2)^{2k+2}}{n}\left[-\frac{1}{2}\log\left(\frac{2}{1+z_0^2}\right) - \frac{\pi}{4} + \arctan(z_0)\right] + \frac{(-1)^{k}(2k+2)^{2k+2}}{n}\left[- G_k(1) + G_k(z_0) -F_{k+1}(1) + F_{k+1}(z_0)\right]$$

For the particular case $k = 2$ ($p=3$ in the question) you asked for this reduces to

$$I_2(1) = \frac{27}{2}\log \left(\frac{z_0^2+1}{2}\right) - \frac{27}{2}(z_0-1)^2 - 28\arctan(z_0) + \frac{29 \pi}{4} \simeq 0.795296 \\ \text{where } z_0 = \frac{1}{3} \left(1+2 \cos \left(\frac{\pi }{9}\right)\right)$$

Finlly since the expression has many terms we should check that it gives the correct result. Here is some code used to check this in Mathematica:

(* Working precision *)
ndigits = 100;

Block[{$MinPrecision = ndigits},

  (* Pick k and n *)
  k = 4; n = 1;

  (* Solve for z0 *)
  z0 = Max[N[x /. Solve[{x^(k + 1) - x^k + n/(k + 1)^(k + 1) == 0, x > 0}, x], ndigits]];

  (* Define the integral *)
  ak = Table[i/(k + 1), {i, 1, k}];
  bk = Table[If[i == k, (i + 1)/k, i/k], {i, 2, k}];
  exact = NIntegrate[ArcTan[HypergeometricPFQ[ak, bk, (x n)/k^k]], {x, 0, 1}, WorkingPrecision -> ndigits];

  (* Our solution *)
  result = If[Mod[k, 2] == 0,

     (* For even k *)
     \[Pi]/2 - ArcTan[z0] + ((-1)^(k/2) (k + 1)^(k + 1))/ n (1/2 Log[2/(1 + z0^2)] - \[Pi]/4 + ArcTan[z0] + Sum[(-1)^i ((1 - z0^(2 i))/(2 i) - (1 - z0^(2 i - 1))/(2 i - 1)), {i, 1, k/2}])


     (* For odd k *)
     \[Pi]/2 - ArcTan[z0] + ((-1)^((k - 1)/2) (k + 1)^(k + 1))/n (-(1/2) Log[2/(1 + z0^2)] - \[Pi]/4 + ArcTan[z0] - Sum[(-1)^i ((1 - z0^(2 i))/(2 i)If[i == (k - 1)/2 + 1, 0, 1] + (1 - z0^(2 i - 1))/(2 i - 1)), {i, 1, (k - 1)/2 + 1}])


  (* Difference between integral and our formula *)
  Print["Difference: ", N[result - exact]];


This demonstrates that we get the same result (to numerical precision) for a wide range of $k$ and $n$.

$^{(1)}$ It's easy to check that this holds to arbitrary precisison (checked to a few hundred digits) at least for the smallest values of $k$. Marty Cohen found this reference discussing a very similar type of property which might be useful if anyone wants to try to derive it (which likely is a very messy affair).

  • $\begingroup$ +1 Thanks for this highly detailed explanation. If I could accept two answers, I would do so. :) $\endgroup$ – Tito Piezas III Jan 4 '17 at 5:07
  • $\begingroup$ Unfortunately, the only justification for that amazing equation (25) is "(M. L. Glasser, pers. comm., Sept. 26, 2003)". So, we do not have a proof, just as assertion. $\endgroup$ – marty cohen Jan 4 '17 at 5:52
  • $\begingroup$ @martycohen, Interesting. Maybe I will post a question about it later. $\endgroup$ – Zaid Alyafeai Jan 4 '17 at 5:55
  • 1
    $\begingroup$ Doing a Google search for "M. L. Glasser hypergeometric functions" results in a number of interesting hits including this one: researchgate.net/profile/Larry_Glasser/publication/… $\endgroup$ – marty cohen Jan 4 '17 at 6:07
  • $\begingroup$ It's more complicated than I though (and far beyond what I'm interested in doing) to derive the formula. To show it one needs to establish $$\frac{1}{1-x^k} = \sum_{n=0}^\infty \frac{\left(\frac{1}{k+1}\right)_n\cdots \left(\frac{k}{k+1}\right)_n}{\left(\frac{2}{k}\right)_n\cdots \left(\frac{k-1}{k}\right)_n\left(\frac{k+1}{k}\right)_n}\frac{x^{nk}(1-x^k)^{nk}(k+1)^{n(k+1)}}{k^{nk} n!}$$ where $(a)_n$ is the falling factorial. $\endgroup$ – Winther Jan 4 '17 at 6:37

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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