Consider the following integral for real $a, b$ such that the square root is real: \begin{equation} I=\int_0^{2 \pi} \sqrt{1-(a+b \sin\phi)^2} d\phi \end{equation} For $a = 0$, the integral is easily expressed in terms of the complete elliptic function of the second kind. Does the integral have a closed form (in terms of special functions) for arbitrary $a$? Mathematica did not yield a useful result, even for specific exact values of $a,b$.
-
2The short answer is yes it does have a solution in terms of elliptic functions plus some elementary parts. To get the full result requires a number of transformations. To start, it is probably best to consider the integral as four parts of width $\pi/2$ as this will allow you to use e.g., $\phi=\arcsin(x)$. This integral is doable but there is a lot of algebra involved. A very good reference for the relevant transformations if you can get a copy is Whittaker and Watson, "A course on modern analysis" ...not very "modern" though as it was written in 1915 :) – Graham Hesketh Apr 27 '14 at 13:20
-
Thanks. It would be very helpful if you (or a potential answerer) could be bit more specific about how to attack the integral. – user111187 Apr 27 '14 at 20:05
1 Answers
Result:
$$\boxed{\displaystyle \mathcal{I}=4\sqrt{\frac{b}{k}}\,\biggl[\mathbf{E}\left(k^2\right)-\left(1-k^2\right)\mathbf{K}\left(k^2\right)+\left(1-k^2\right)\Pi\left(c^{-2}|k^2\right)\biggr]} \tag{$\heartsuit$}$$
where I follow Mathematica conventions for arguments of $\mathsf{EllipticE}$, $\mathsf{EllipticK}$ and $\mathsf{EllipticPi}$, and the parameters $k$ and $c$ are defined below by (0) and (2), respectively.
Derivation:
$\square$ The underlying elliptic curve $\mathcal{C}$ is described by the equation $$y^2=\left(1-z^2\right)\left(1-(a+bz)^2\right).$$ Topologically this is a torus realized as a two-sheeted covering of $\mathbb{P}^1$ branched at $4$ points $z_{1\ldots4}=\pm1,\frac{\pm1-a}{b}$. These can be put to $\pm1,\pm k^{-1}$ by a fractional linear transformation. Such transformations preserve the anharmonic ratio, which allows to identify $$\frac{z_{21}z_{43}}{z_{31}z_{42}}=\frac{4b}{a^2-(1-b)^2}=-\frac{4k}{(1-k)^2}.$$ Choose a solution $$k=\frac{1+b^2-a^2-\sqrt{(1+b^2-a^2)^2-4b^2}}{2b}.\tag{0}$$ The relevant fractional linear transformation bringing $\mathcal{C}$ to the standard Legendre form is thus $$z=\frac{a w+(k-b)}{(k-b) w+a}.$$
After these preliminary remarks, we can implement the corresponding change of variables in the integral we are interested in: \begin{align} \mathcal{I}=&\int_0^{2\pi}\sqrt{1-(a+b\sin\phi)^2}\,d\phi=\\=& 2\int_{-\frac{\pi}{2}}^{\frac{\pi}{2}}\sqrt{1-(a+b\sin\phi)^2}\,d\phi=\\ =&2\int_{-1}^1 \sqrt{\frac{1-(a+bz)^2}{1-z^2}}dz=\\=&2\left[\frac{a^2}{(b-k)^2}-1\right]\sqrt{\frac{b}{k}}\int_{-1}^1\sqrt{\frac{1-k^2w^2}{1-w^2}}\frac{dw}{\left(w-\frac{a}{b-k}\right)^2}=\\ =&2\left(c^2-1\right)\sqrt{\frac{b}{k}}\int_{-1}^1\frac{1-k^2w^2}{\left(w-c\right)^2}\frac{dw}{\lambda},\tag{1} \end{align} with $\lambda=\sqrt{(1-w^2)(1-k^2w^2)}$ and $$c=\frac{a}{b-k}.\tag{2}$$ It is possible to reduce the last integral to a combination of integrals of three types: $$\int\frac{dw}{\lambda},\qquad \int\frac{w^2dw}{\lambda},\qquad \int\frac{dw}{(w-c)\lambda},\tag{3}$$ producing elliptic integrals of the 1st, 2nd and 3rd type, respectively. The idea is to write \begin{align} \frac{1-k^2w^2}{\left(w-c\right)^2}\frac{1}{\lambda}=\left[\alpha+\frac{\beta}{w-c}+\frac{\gamma}{(w-c)^2}\right]\frac{1}{\lambda} \end{align} The $\alpha$- and $\beta$-term give elliptic integrals of the 1st and 3rd type. To reduce the $\gamma$-term to integrals of type (3), compare it with the derivative \begin{align}\frac{d}{dw}\frac{\lambda}{w-c}=-\frac{(c^2-1)(c^2k^2-1)}{(w-c)^2\lambda} -\frac{c(c^2k^2-1)+ck^2(c^2-1)}{(w-c)\lambda}+\frac{k^2(w^2-c^2)}{\lambda}. \end{align}
Altogether this allows us to write $$\frac{1-k^2w^2}{\left(w-c\right)^2}\frac{1}{\lambda}=\frac{1}{c^2-1}\left[\frac{d}{dw}\frac{\lambda}{w-c}-\frac{c(1-k^2)}{(w-c)\lambda}+\frac{k^2(1-w^2)}{\lambda}\right],$$ which in turn implies $$\mathcal{I}=2\sqrt{\frac{b}{k}}\int_{-1}^1\left[-\frac{c(1-k^2)}{(w-c)\lambda}+\frac{k^2(1-w^2)}{\lambda}\right]dw.$$ After some straightforward algebra, this finally gives the above result ($\heartsuit$). $\blacksquare$
- 54,260
-
1Excellent answer, thank you very much. It agrees numerically for generic values of $a$ and $b$, but Mathematica throws an error at $a=b=1/2$, even though the integral is well-defined. It does give the correct numerical result when taking, e.g, $b = 0.4999999999$. – user111187 Apr 29 '14 at 11:09
-
@user111187 For $a=b=1/2$ we have $k=1$, I guess this is the source of the problem for Mathematica. Actually the answer was obtained assuming that $a$ and $b$ are small enough (note that I tried to adjust the notation so that $k\rightarrow b$ as $a\rightarrow 0$) and for larger values should be understood in the sense of analytic continuation. I think that if you set $a=b=s$ and then let $s\rightarrow \frac12^-$, it will get correct finite value. – Start wearing purple Apr 29 '14 at 11:28