20

Is there a closed form for $$I_n =\int_{0}^{\pi} \frac{\phi}{(1 - r \cos\phi)^n} \,{\rm d}\phi $$ for $\left\vert\,r\,\right\vert < 1$ real and $n > 0$ integer ? The solution to this integral would give a closed form solution for this integral, which describes the interaction energy of vector resonant relaxation in astrophysical dynamics.

Using Mathematica and analytic methods I have derived the following result for $n=\{1,2,3,4,5,6\}$: $$\tag{1} I_n = -\frac{a_n r}{s^{2n-2}} + \frac{b_n}{s^{2n-1}}\left[\chi_2(q) +\left(-\frac{c_n}{b_n}s + {\rm arctanh}(q) \right){\rm arctanh}(r) \right]$$ where $s=\sqrt{1-r^2}$, $q=\sqrt{(1-r)/(1+r)}$, $\chi_2(x)$ is the Legendre chi function, $a_n$, $b_n$, and $c_n$ are constants given by \begin{align} a_1 &= 0, \quad b_1 = 4, \qquad c_1 = 0,\\ a_2 &= 0, \quad b_2 = 4, \qquad c_2 = 2,\\ a_3 &= 1, \quad b_3 = 4 + 2r^2, \qquad c_3 = 3,\\ a_4 &= \frac{7}{3}, \quad b_4 = 4 + 6r^2, \qquad c_4= \frac{11}{3} + \frac{4}{3}r^2,\\ a_5 &= \frac{23}{6}+\frac{11}{12}r^2, \quad b_5 = 4 + 12 r^2 + \frac{3}{2}r^4, \qquad c_5=\frac{25}{6} + \frac{55}{12}r^2,\\ a_6 &=\frac{163}{30} + \frac{47}{12}r^2, \quad b_6 = 4 + 20 r^2 + \frac{15}{2}r^4, \qquad c_6=\frac{137}{30} + \frac{607}{60}r^2 + \frac{16}{15}r^4. \end{align} Is the integral $I_n$ in the general closed form given by (1) for all $n$? If so, what are the constants $a_n$, $b_n$, and $c_n$?

bkocsis
  • 1,238
  • 2
    @robjohn The same thing could have been mentioned in a more polite manner. There's nothing wrong in your statement but it came off a bit rude. No offense! – Ahaan S. Rungta Nov 26 '13 at 20:43
  • 7
    @AhaanRungta: I did not mean it to. I said Please and Thanks. What would have been a more polite way? – robjohn Nov 26 '13 at 20:44
  • 1
    a.) Please X. Thanks.

    b.) Could you please x so that Y? Edited this time. Thank you. <- which do you think sounds better?

    – Ahaan S. Rungta Nov 26 '13 at 20:46
  • @robjohn No problem, if the equation didn't show up in the title for some reason, I have added it to the main body. – bkocsis Nov 26 '13 at 20:46
  • Hypergeometric functions first come into my mind... – Sangchul Lee Nov 26 '13 at 20:46
  • @AhaanRungta: The link provided a much more comprehensive reason why. I didn't think I needed to repeat it. – robjohn Nov 26 '13 at 20:47
  • @bkocsis: Don't worry; it wasn't your doing. Someone else edited the title to use displaystyle. – robjohn Nov 26 '13 at 20:54
  • @sos440: It is no need to call for the hypergeometric functions as $n$ in this question is an integer. – doraemonpaul Nov 27 '13 at 07:01
  • You could improve this question by including more context. In particular, what is the interest in this integral? There is a general description about how to write questions on this site at http://meta.math.stackexchange.com/questions/9959/how-to-ask-a-good-question – Carl Mummert Nov 27 '13 at 13:25
  • This integral should has closed form. You can just apply the Weierstrass substitution, the only thing is how the tedious rational function integral you should face. – doraemonpaul Nov 27 '13 at 06:56
  • Needs more detail to be considered an answer and not just a comment, I think. – Bennett Gardiner Nov 27 '13 at 09:52
  • While not a complete answer, it is a useful comment. The substitution $t=\tan(x/2)$, $x=2{\rm arctan,}t$, gives $$(1−r)^{−n}\int_0^{\infty}\frac{ 4 (1+t^2)^{n−1} {\rm arctan,}t}{(1+q^2t^2)^n} dt$$ where $q=(1+r)/(1−r)$. While Mathematica 9.0 cannot solve this for any arbitrary positive integer $n$, it does produce closed form solutions for $n={1,2,3,4,5,6}$ using ${\rm arctanh,}(q)$, ${\rm LerchPhi,}(q^2,2,k/2)$, and ${\rm PolyLog,}(2,\pm q)$ functions where $k={1,3,\dots,2n−1}$. Perhaps there is a closed form solution for arbitrary $n$ using these functions. – bkocsis Nov 27 '13 at 17:10
  • 1
    @CarlMummert: I added the context to the question. – bkocsis Nov 27 '13 at 20:49
  • The same question at MO: http://mathoverflow.net/questions/150539/closed-form-for-int-0-pi-frac-phi1-r-cos-phin-d-phi – Martin Sleziak Dec 02 '13 at 14:03

3 Answers3

7

First lets make the integral from $-\pi/2$ to $\pi/2$. Set $x = \phi-\pi/2$. We then get $$I = \int_{-\pi/2}^{\pi/2}(1+r\sin(x))^{-n}(x+\pi/2)dx$$ Now recall that $$(1+r\sin(x))^{-n} = \sum_{k=0}^{\infty} (-1)^k \dbinom{n+k}k r^k \sin^k(x)$$ Hence, $$I = \sum_{k=0}^{\infty}(-r)^k \dbinom{n+k}k \int_{-\pi/2}^{\pi/2} (x+\pi/2)\sin^k(x)dx$$ Now from here, we can obtain $$\int_{-\pi/2}^{\pi/2} \sin^k(x) dx \text{ and }\int_{-\pi/2}^{\pi/2} x\sin^k(x) dx$$ Lets call them $I_k$ and $J_k$ respectively. Hence, $$I = \dfrac{\pi}2\sum_{k=0}^{\infty}r^{2k} \dbinom{n+2k}{2k} I_{2k} - \sum_{k=0}^{\infty}r^{2k+1} \dbinom{n+2k+1}{2k+1} J_{2k+1} \tag{$\star$}$$


Expression for $J_{2k+1}$ and convergence of the above summation:

We have $$J_{2k+1} = -\int_{-\pi/2}^{\pi/2}x \sin^{2k}(x) d(\cos(x)) = \int_{-\pi/2}^{\pi/2} \cos(x) \sin^{2k}(x) dx + 2k \int_{-\pi/2}^{\pi/2} x \cos^2(x) \sin^{2k-1}(x) dx$$ $$\int_{-\pi/2}^{\pi/2} \cos(x) \sin^{2k}(x) dx = \int_{-1}^1 t^{2k} dt = \dfrac2{2k+1}$$ $$\int_{-\pi/2}^{\pi/2} x \cos^2(x) \sin^{2k-1}(x) dx = J_{2k-1} - J_{2k+1}$$ Hence, $$(2k+1)J_{2k+1} = \dfrac2{2k+1} + 2k J_{2k-1}$$ Using this recurrence you can obtain $J_{2k+1}$. Also, it is easy to show that $\left \vert J_{2k+1} \right \vert \leq 2$ using induction.

Similarly, $I_{2k} = \dfrac1{4^k} \dbinom{2k}k \pi \leq \pi$.

Now, for $\vert r \vert < 1$, we have $\displaystyle \sum_{k=0}^{\infty} \dbinom{n+2k}{2k} r^{2k}$ and $\displaystyle \sum_{k=0}^{\infty} \dbinom{n+2k+1}{2k+1} r^{2k}$ converges.

This ensures $\star$ makes perfect sense.

  • I was looking for a closed form expression, i.e. a finite sum of known functions. See edited version of the question. – bkocsis Dec 01 '13 at 17:41
  • @bkocsis I don't understand what you mean by closed form. The above is a closed form in my opinion. If you get the answer as $\sin(k)$ or $\cos(k)$ or $\Gamma(k)$ or $\arctan(k)$, etc, do you think that is closed form? All these are infinite series and not a closed form. By truncating the above sum, you can get an arbitrary accurate answer. –  Dec 01 '13 at 17:48
  • 1
    @bkocsis Also, can you remove the down-vote. My answer is a perfectly valid answer. –  Dec 01 '13 at 17:55
  • @bkocsis You are showing a wiki article to mean what a closed form is? I assume that you are interested in calculating the value of the integral for a fixed $r$ and $n$. If so, the infinite series above is a perfectly valid answer. –  Dec 01 '13 at 23:17
  • The beauty of closed form expressions is that the parameter dependencies become tractable, and that they are often calculable quickly. Most CPUs and GPUs can calculate many transcendental functions extremely efficiently. Closed form results can speed up simulations very significantly. If you can show that your sums converge for $|r|<1$ and give $J_{2k+1}$ specifically, I can remove the down vote. – bkocsis Dec 02 '13 at 00:14
  • @bkocsis Added. Also, if you need, I can send you a robust implementation accurate up-to $15$ digits. –  Dec 02 '13 at 01:29
6

The integral can be evaluated by realizing that it is a derivative with respect to $R=1/r$, i.e. $$I_n =\int_{0}^{\pi} \frac{\phi}{(1 - r \cos\phi)^n} \,{\rm d}\phi = R^n \int_{0}^{\pi} \frac{\phi}{(R - \cos\phi)^n} \,{\rm d}\phi = \frac{(-1)^{n-1} R^n}{(n-1)!} \frac{d^{n-1}}{dR^{n-1}}\int_{0}^{\pi} \frac{\phi}{R - \cos\phi} \,{\rm d}\phi. $$ This integral can be evaluated with a Weierstrass substitution $t = \tan (\phi/2)$, $$F(R) = \int_{0}^{\pi} \frac{\phi}{R - \cos\phi} \,{\rm d}\phi = \frac{4r}{1+r}\int_0^{\infty} \frac{\arctan t}{q^2 + t^2} dt = \frac{4}{\sqrt{R^2-1}}\int_0^{\pi/2} \arctan \left(q \tan y \right) dy = \frac{4}{\sqrt{R^2-1}}\left[\chi_2(q) - {\rm arctanh}(q) \ln(q) \right] =\frac{4}{\sqrt{R^2-1}}\left[\chi_2(q) +\frac{1}{2} {\rm arccosh}(R) {\rm arctanh}(R) \right],$$ where $q=\sqrt{(1-r)/(1+r)}$ and $\chi_2(x)$ is the Legendre chi function. We can now derive the integral by taking derivatives of this function $$I_n = \frac{(-1)^{n-1} R^n}{(n-1)!} \frac{d^{n-1}}{dR^{n-1}}F(R).$$ Note that ${\rm arccosh\,}R=2{\rm arctanh\,}q$, ${\rm arccoth\,}R=\ln q$, and $$\frac{d}{dx}\chi_2(x) = \frac{{\rm arctanh}(x)}{x} \quad{\rm so\quad} \frac{d}{dR}\chi_2(q) = \frac{{\rm arctanh}(q)}{R^2 - 1} = \frac{1}{2}\frac{{\rm arccosh}(R)}{R^2 - 1} $$ $$\frac{d}{dR} {\rm arccosh}(R) = \frac{1}{\sqrt{R^2 - 1}} \quad{\rm and\quad} \frac{d}{dR} {\rm arccoth}(R) = \frac{-1}{R^2 - 1} = \frac{1}{2}\left(\frac{1}{R+1} - \frac{1}{R-1}\right).$$ The derivatives for $n\geq 3$ simplify to $$I_n = \frac{R^n}{S^{2n-2}}\left\{ \frac{4 A_{n-1}}{S}\left[\chi_2(q) - {\rm arctanh(q)\ln q}\right]\right. \\ \left.+2\sum_{k=0}^{n-2} \frac{A_k A_{n-k-2}}{k+1} \ln q + \sum_{j=0}^{n-3}\sum_{k=0}^{n-3-j}\frac{A_k A_{n-3-j-k}}{n-1-k} \frac{(R-1)^{1+j}-(R+1)^{1+j}}{1+j} \right\} $$ where $S=\sqrt{R^2-1}$ and $A_n$ is a polynomial of $R$ defined as $$A_n = \frac{(-1)^n}{n!}\left(R^2-1\right)^{n+(1/2)}\frac{d^{n}}{dR^{n}} \left(R^2-1\right)^{-1/2}\\ =\left(\frac{R+1}{4}\right)^n\sum_{s=0}^{n}\frac{(2s)!}{(s!)^2}\frac{[2(n-s)]!}{[(n-s)!]^2}q^{2s}.$$ The coefficients are related to Legendre polynomials $P_{2n}(0)=(-1)^n(2n)!/[4^n (n!)^2]$. The result simplifies to $$ I_n=\left(\frac{R}{R-1}\right)^n \frac{q^2}{4^{n-2}}\left\{ a(q) [\chi_2(q) - {\rm arctanh(q)}\ln q] +b(q)\ln q + c(q)\right\}$$ where $$ a(q)=\sum_{K=0}^{n-1}q^{2K-1} \frac{(2K)!}{(K!)^2} \frac{[2(n-1-K)]!}{[(n-1-K)!]^2}\,, $$ $$ b(q)=\sum_{K=0}^{n-2}q^{2K} \sum_{s=0}^{K}\frac{(2s)!}{(s!)^2}\frac{[2(K-s)]!}{[(K-s)!]^2} \sum_{j=K}^{n-2}\frac{[2(j-K)]!}{[(j-K)!])^2}\frac{[2(n-2-j)]!}{[(n-2-j)!]^2}\frac{2}{j-s+1}\,, $$ $$ c(q)=\sum_{K=0}^{n-3}q^{2K+2} \sum_{J=0}^{K} \frac{4^{K-J+1}}{1+K-J} \sum_{s=0}^{J}\frac{(2s)!}{(s!)^2}\frac{[2(J-s)]!}{[(J-s)!]^2}\\ \quad\times \sum_{j=K}^{n-3}\frac{[2(j-K)]!}{[(j-K)!])^2}\frac{[2(n-3-j)]!}{[(n-3-j)!]^2}\frac{1}{n-1-j+s+K-J}\\ \quad+\sum_{K=0}^{n-3}q^{2K} \sum_{J=K}^{L-3} \frac{-4^{J-K+1}}{1+J-K}\sum_{s=0}^{K}\frac{(2s)!}{(s!)^2}\frac{[2(K-s)]!}{[(K-s)!]^2}\\ \quad\times \sum_{j=J}^{n-3}\frac{[2(j-J)]!}{[(j-J)!])^2}\frac{[2(n-3-j)]!}{[(n-3-j)!]^2}\frac{1}{n-1-j+s+J-K}\,.$$

bkocsis
  • 1,238
3

Not a full answer, but maybe interesting:

$I_{n}(r)$ obeys a recursion relation: $$ I_{n+1}(r)=I_{n}(r)+\frac{r}{n} \frac{\partial}{\partial r}I_{n}(r) $$

A large $n$ approximation is: $$ I_{n}(r)\stackrel{n \rightarrow \infty}{\sim} \int_{0}^{\pi}\phi \exp(n\, r \cos \phi)\mathrm{d}\phi, $$ which shows that for large $n$ the integral depends only on the product $n\,r$. Interestingly, also the approximation seems not to be solvable in closed form.

For the approximation I used: $$ \left(1 - \frac{z}{n}\right)^{n} \stackrel{n \rightarrow \infty}{\sim} e^{- z} $$