Q: How to evaluate the following integral? $$ PV\int_k^1\frac{dt}{(x^2-t^2)\sqrt{(t^2-k^2)(1-t^2)}} $$ where $k<x<1$. For the sake of definiteness we may assume $x=0.8, k=0.5$.

Attempt: The substitution $t^2=k^2\cos^2\theta+\sin^2\theta$ can be used to transform the integral into $$ {1\over x^2-1}\int_0^{\pi\over2}\frac{d\theta}{\left(1-\frac{1-k^2}{1-x^2}\sin^2\theta\right)\sqrt{1-(1-k^2)\sin^2\theta}} $$ which can be expressed as the complete elliptic integral of third kind $$ {1\over x^2-1}\Pi\left(\frac{1-k^2}{1-x^2},\sqrt{1-k^2}\right) $$ if $$\frac{1-k^2}{1-x^2}<1\iff x^2<k^2$$ in which case there would not be any singularity in the denominator. But in our case this is not true and the integrand is singular at $$\theta=\arcsin\left(\sqrt{\frac{1-x^2}{1-k^2}}\right)$$
Can someone show me the right contour for this integral and the nuances that come with it?


I do not know the right contour for this integral but there are no general problems to compute it for your values. With $$f(k,x) =\int_k^1\frac{dt}{(x^2-t^2)\sqrt{(t^2-k^2)(1-t^2)}}$$ using Maple integration I get $f(0.5, 0.8)=1.78709921080526937$. If you use the Maple form of $\Pi$ $$g(k,x) = \frac{1}{x^2-1}\mathtt{EllipticPi}(\frac{1-k^2}{1-x^2}, \sqrt{1-k^2})$$ you get also $g(0.5, 0.8)=1.78709921080526937$

My own special function library use the Bulirsch function cel see http://dlmf.nist.gov/19.2#iii $$ \Pi(\nu,k) = \mathrm{cel}(\sqrt{1-k^2},1-\nu,1,1)$$ with a port of the Algol procedure including the Cauchy principal value handling for $\nu > 1$, from R. Bulirsch, Numerical Calculation of Elliptic Integrals and Elliptic Functions, part III. Numerische Mathematik 13, 305-315, 1969 (available from http://www.digizeitschriften.de/en/dms/toc/?PPN=PPN362160546_0013)

An alternative implementation with the Carlson form from http://dlmf.nist.gov/19.25 (where you find an explicit case for the Cauchy principal value) is

$$\Pi(\nu,k) = R_F(0,{k^{'}}^{2},1) + \frac{\nu}{3} R_J(0,{k^{'}}^{2},1,1-\nu), \quad {k^{'}}^{2}= 1-k^2 $$

Both library versions give $1.78709921080526937\,$ for your example values.

  • $\begingroup$ Awesome answer. I was using mathematica and it returned imaginary numbers. $\endgroup$ – Jack's wasted life Sep 21 '16 at 8:39
  • $\begingroup$ It seems, that you can use the real part of the result from Wolfram Alpha (Mathematica). With the input EllipticPi[0.75/0.36,0.75]/(-0.36) with 18 digits WA returns $1.78709921080526937+5.24018157944748161\times i.$ Note that WA uses $m=k^2$ for the integrals. $\endgroup$ – gammatester Sep 21 '16 at 10:01
  • $\begingroup$ I added another answer, which avoids the sticky issues of how $\Pi$ is defined for $n > 1$. $\endgroup$ – Maxim May 24 '18 at 15:30

Using the parameter notation, $$PV \int_k^1 \frac {dt} {(x^2 - t^2) \sqrt {(t^2 - k^2) (1 - t^2)}} =\\ \frac 1 {1 - x^2} \left( \Pi(1 - x^2, 1 - k^2) - K(1 - k^2) \right), \\ 0 < k < x < 1.$$ This is independent of how $K$ and $\Pi$ are extended to arguments outside $(0, 1)$. Can be derived in a similar way to this.


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.