Define the function $\mathcal{I}:\left(0,1\right)\rightarrow\mathbb{R}$ via the definite integral


Question: Given $p\in\left(0,1\right)$, is there a (relatively) simple way to express $\mathcal{I}{\left(p\right)}$ in terms of elliptic integrals in Legendre normal form? Can $\mathcal{I}{\left(p\right)}$ possibly be written as a simple scalar multiple of a complete elliptic integral of the first kind?

Recall that for amplitude $\theta\in\left(0,\frac{\pi}{2}\right)$ and elliptic modulus $\kappa\in\left(0,1\right)$, the incomplete elliptic integral of the first kind $F{\left(\theta,\kappa\right)}$ is defined as



I have noticed that of the questions on this site related to proving special closed form values of the Gauss hypergeometric function, ${_2F_1}{\left(a,b;c;z\right)}$, an unusually large number of the more difficult ones can be shown equivalent to proving the particular hypergeometric function with parameters $a=\frac12$, $b=\frac13$, and $c=\frac56$, i.e., ${_2F_1}{\left(\frac12,\frac13;\frac56;z\right)}$, takes some special (usually algebraic) value for some particular (usually rational) value of $z$, where $0<z<1$. So in the interest of generalization, I wanted to see how close one could get to obtaining a closed form for $f(z):={_2F_1}{\left(\frac12,\frac13;\frac56;z\right)}$ for arbitrary values of $z\in(0,1)$.

Using Euler's integral representation for the Gauss hypergeometric function with the assumption that $0<z<1$, we have:

$$\begin{align} f(z) &={_2F_1}{\left(\frac12,\frac13;\frac56;z\right)}\\ &=\frac{1}{\operatorname{B}{\left(\frac13,\frac12\right)}}\int_{0}^{1}t^{-\frac23}(1-t)^{-\frac12}(1-zt)^{-\frac12}\mathrm{d}t\\ &=\frac{3}{\operatorname{B}{\left(\frac13,\frac12\right)}}\int_{0}^{1}\frac{\mathrm{d}x}{\sqrt{(1-x^3)(1-zx^3)}},\tag{3}\\ \end{align}$$

where the beta function factor can alternatively be written in terms of the gamma function as,


The integral factor in the last line of $(3)$ is already superficially quite similar to the following integral representation of the complete elliptic integral of the first kind:


Moreover, the last line of $(3)$ implies that in order for $f(z)$ to take an algebraic value, the integral factor will be the product of an algebraic number $A$ with an elliptic integral singular value:


This made me suspect that the integral $\int_{0}^{1}\frac{\mathrm{d}x}{\sqrt{(1-x^3)(1-zx^3)}}=\mathcal{I}{(z^{1/6})}$ might be capable of being written as a linear combination of complete elliptic integrals in general.

  • $\begingroup$ Do you mind if I ask how you came up with the substitution $x=\frac{1}{a}(z-\sqrt{z^2-1})$ ? Thanks. $\endgroup$
    – M.N.C.E.
    Oct 25 '14 at 9:18
  • $\begingroup$ @M.N.C.E. See my question about integrals of reciprocal square-roots of 6th degree polynomials here. $\endgroup$
    – David H
    Oct 25 '14 at 9:27
  • $\begingroup$ A route I'll try to go after is that $_2 F_1(a,b;c;z)$ satisfies an ODE. So we can transfer the problem from integration to a boundary value problem. That's hardly a panacea, but it offers some possibilities. $\endgroup$ Oct 27 '14 at 23:41
  • 1
    $\begingroup$ For reference, and to to clarify my last comment: $y={_2F_1}(\frac12,\frac13;\frac56;z)$ satisfies the hypergeometric ODE $$6z(1-z)y''(z)+(5-11z)y'(z)-y(z)=0.$$ If we work in terms of $\alpha$ via $z=\alpha^6$, then the ODE becomes $$216(\alpha^6-1) \alpha \,y''(\alpha)+6(41\alpha^6-35)\,y'(\alpha)+\alpha^5\, y(\alpha)=0.$$ (This has the advantage that both solutions are analytic near 0, whereas the first had a second solution with leading term $z^{1/6}$.) $\endgroup$ Oct 28 '14 at 3:18
  • 2
    $\begingroup$ This is very closely related to this question. $\endgroup$
    – Kirill
    Oct 28 '14 at 14:59

General Procedure:

I will give an outline of a possible procedure (that I learned from a calculus book some time ago) to solve integrals of the form $$\mathcal{J}=\int\frac{{\rm d}x}{\sqrt{(x^2+ax+b)(x^2+cx+d)}}$$ First, let $x=t+\lambda$. \begin{align} \mathcal{J}=\int\frac{{\rm d}t}{\sqrt{(t^2+(2\lambda+a)t+\lambda^2+\lambda a+b)(t^2+(2\lambda+c)t+\lambda^2+\lambda c+d)}} \end{align} Next, let $\lambda=\dfrac{d-b}{a-c}$, then make the substitution $\displaystyle t=\mu v$ where $\displaystyle \mu=\sqrt{\lambda^2+\frac{ad-bc}{a-c}}$. $$\mathcal{J}=\frac{1}{\mu}\int\frac{{\rm d}v}{\sqrt{\left(v^2+Av+1\right)\left(v^2+Cv+1\right)}}$$ where $\displaystyle A=\frac{2\lambda+a}{\mu}$ and $\displaystyle C=\frac{2\lambda+c}{\mu}$. Follow this up with the substitution $\displaystyle v=\frac{1+w}{1-w}$. $$\mathcal{J}=\frac{2}{\mu}\int\frac{{\rm d}w}{\sqrt{((2-A)w^2+(2+A))((2-C)w^2+(2+C))}}$$ By applying the appropriate trigonometric substitution, it can be seen that $\mathcal{J}$ can be reduced to incomplete elliptic integrals.


I will define the following parameters to save space and typing. \begin{align} \alpha_1=&\ \frac{1}{\sqrt{\left(a+1\right)^2\left(3a^2+2\sqrt{3}\sqrt{a^4+a^2+1}+3\right)}}\\ \alpha_2=&\ \frac{1}{\sqrt{\left(a-1\right)^2\left(3a^2+2\sqrt{3}\sqrt{a^4+a^2+1}+3\right)}}\\ \beta_1=&\ \frac{\sqrt{3}a^2-\sqrt{a^4+a^2+1}+\sqrt{3}a+\sqrt{3}}{\sqrt{3}a^2+\sqrt{a^4+a^2+1}+\sqrt{3}a+\sqrt{3}}\\ \beta_2=&\ \frac{\sqrt{3}a^2-\sqrt{a^4+a^2+1}-\sqrt{3}a+\sqrt{3}}{\sqrt{3}a^2+\sqrt{a^4+a^2+1}-\sqrt{3}a+\sqrt{3}}\\ \xi_1=&\ \frac{\sqrt{a^4+a^2+1}-\sqrt{3}a}{\sqrt{a^4+a^2+1}+\sqrt{3}a}\\ \xi_2=&\ \frac{\sqrt{a^4+a^2+1}+\sqrt{3}a}{\sqrt{a^4+a^2+1}-\sqrt{3}a} \end{align}

Evaluation of $\ \mathcal{I}(a)$

It was derived by the OP that \begin{align} \mathcal{I}(a) =&\underbrace{\frac{1}{4a}\int^\infty_\frac{a^2+1}{2a}\frac{{\rm d}x}{\sqrt{\left(x^2-\frac{(a-1)^2}{2a}x-\frac{a^2+1}{2a}\right)\left(x^2+\frac{a^2+1}{2a}x+\frac{a^4-a^2+1}{4a^2}\right)}}}_{\mathcal{I}_1}\\ &+\underbrace{\frac{1}{4a}\int^\infty_\frac{a^2+1}{2a}\frac{{\rm d}x}{\sqrt{\left(x^2-\frac{(a+1)^2}{2a}x+\frac{a^2+1}{2a}\right)\left(x^2+\frac{a^2+1}{2a}x+\frac{a^4-a^2+1}{4a^2}\right)}}}_{\mathcal{I}_2} \end{align} Applying the aforementioned procedure and using Wolfram Alpha to do all the algebra, we get \begin{align} \mathcal{I}_1 =&\frac{1}{4a}\int^\infty_\frac{a^2+1}{2a}\frac{{\rm d}x}{\sqrt{\left(x^2-\frac{(a-1)^2}{2a}x-\frac{a^2+1}{2a}\right)\left(x^2+\frac{a^2+1}{2a}x+\frac{a^4-a^2+1}{4a^2}\right)}}\\ =&\frac{1}{\sqrt{3}\sqrt{a^4+a^2+1}}\int^\infty_\frac{\sqrt{3}(a^2+a+1)}{\sqrt{a^4+a^2+1}}\frac{{\rm d}x}{\sqrt{\left(x^2-\frac{2(2a^2+a+2)}{\sqrt{3}\sqrt{a^4+a^2+1}}x+1\right)\left(x^2-\frac{2\sqrt{3}\ a}{\sqrt{a^4+a^2+1}}+1\right)}}\\ =&\alpha_1\int^1_{\beta_1}\frac{{\rm d}x}{\sqrt{\left(x^2-\beta_1^2\right)\left(x^2+\xi_1\right)}} =\alpha_1\int^{\arccos{\beta_1}}_{0}\frac{\sec{\varphi}\tan{\varphi}}{\sqrt{(\sec^2{\varphi}-1)(\beta_1^2\sec^2{\varphi}+\xi_1)}}{\rm d}\varphi\\ =&\alpha_1\int^{\arccos{\beta_1}}_{0}\frac{{\rm d}\varphi}{\sqrt{\beta_1^2+\xi_1\cos^2{\varphi}}} =\alpha_1\int^{\arccos{\beta_1}}_{0}\frac{{\rm d}\varphi}{\sqrt{(\beta_1^2+\xi_1)-\xi_1\sin^2{\varphi}}}\\ =&\frac{\alpha_1}{\sqrt{\beta_1^2+\xi_1}}\int^{\arccos{\beta_1}}_{0}\frac{{\rm d}\varphi}{\sqrt{1-\frac{\xi_1}{\beta_1^2+\xi_1}\sin^2{\varphi}}} =\color{red}{\frac{\alpha_1}{\sqrt{\beta_1^2+\xi_1}}\mathbf{F}\left(\arccos{\beta_1},\sqrt{\frac{\xi_1}{\beta_1^2+\xi_1}}\right)} \end{align} for $\mathcal{I}_1$, and we obtain (I omit the steps of trigonometric manipulations since they are the same) \begin{align} \mathcal{I}_2 =&\frac{1}{4a}\int^\infty_\frac{a^2+1}{2a}\frac{{\rm d}x}{\sqrt{\left(x^2-\frac{(a+1)^2}{2a}x+\frac{a^2+1}{2a}\right)\left(x^2+\frac{a^2+1}{2a}x+\frac{a^4-a^2+1}{4a^2}\right)}}\\ =&\frac{1}{\sqrt{3}\sqrt{a^4+a^2+1}}\int^\infty_{\frac{\sqrt{3}(a^2-a+1)}{\sqrt{a^4+a^2+1}}}\frac{{\rm d}x}{\sqrt{\left(x^2-\frac{2(2a^2-a+2)}{\sqrt{3}\sqrt{a^4+a^2+1}}x+1\right)\left(x^2+\frac{2\sqrt{3}\ a}{\sqrt{a^4+a^2+1}}x+1\right)}}\\ =&\alpha_2\int^1_{\beta_2}\frac{{\rm d}x}{\sqrt{\left(x^2-\beta_2^2\right)\left(x^2+\xi_2\right)}}=\color{blue}{\frac{\alpha_2}{\sqrt{\beta_2^2+\xi_2}}\mathbf{F}\left(\arccos{\beta_2},\sqrt{\frac{\xi_2}{\beta_2^2+\xi_2}}\right)} \end{align} Therefore $$\mathcal{I}(a)=\frac{\alpha_1}{\sqrt{\beta_1^2+\xi_1}}\mathbf{F}\left(\arccos{\beta_1},\sqrt{\frac{\xi_1}{\beta_1^2+\xi_1}}\right)+\frac{\alpha_2}{\sqrt{\beta_2^2+\xi_2}}\mathbf{F}\left(\arccos{\beta_2},\sqrt{\frac{\xi_2}{\beta_2^2+\xi_2}}\right)$$ I have substituted in some random values of $a$ into $\mathcal{I}_1$ and it seems to match numerically. However, I have not verified $\mathcal{I}_2$ yet so there might be a mistake.

  • $\begingroup$ I find your answer useful because it explains how to prove Prop.3.145(1) of Gradshteyn, which I didn't previously know how to do, so +1). However, I don't think it's particularly relevant to the problem at hand. Do you have any thoughts about the conjecture I made at the bottom of my post? $\endgroup$
    – David H
    Oct 25 '14 at 13:10
  • $\begingroup$ @DavidH As of now, I do not have an approach in mind to verify your conjecture, sorry for that. However, in the event that I do find a viable method to prove or disprove your conjecture, I would post it (that is assuming that someone else doesn't post their approach before I do). $\endgroup$
    – M.N.C.E.
    Oct 25 '14 at 13:18
  • $\begingroup$ FYI, the conjecture appears to be false. Apparently I should have tested more values. =p $\endgroup$
    – David H
    Oct 25 '14 at 18:03
  • $\begingroup$ @DavidH It is quite a pity that your conjecture is false, it would have saved me alot of work if it were true :P. In any case, I have edited my answer slightly and may I ask if it suits the question requirements better this time round? I will probably check my answer for errors only tomorrow. $\endgroup$
    – M.N.C.E.
    Oct 28 '14 at 12:57

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.