By stressing my manipulative powers and a bit of help from Mathematica I was able to show that the triple integral

$$\frac1{8\pi}\int_0^{2\pi}\int_0^\pi\int_0^\pi\,(2\cos\theta\cos\varphi-\sin\theta\sin\varphi\cos\psi)^{2k}\sin\theta\sin\varphi \,\mathrm{d}\theta\mathrm{d}\varphi\mathrm{d}\psi $$

where k is a positive integer is equivalent to


by making use of the binomial expansion and then using a hypergeometric identity that degenerates to a ratio of gamma functions.

For some reason I think there was no need for Mathematica to invoke the Gauss hypergeometric function ${}_2 F_1\left(a,b;c;z\right)$ in performing the triple integration, so I'm wondering if there might be a way to show that this identity holds, using nothing more complicated than a gamma function.

  • 3
    $\begingroup$ The standard thing to do is to write everything in terms of complex exponentials. This should be completely elementary. $\endgroup$ – Qiaochu Yuan Aug 16 '10 at 23:35
  • $\begingroup$ Somehow, doing that replacement after performing the binomial expansion resulted in an even more hellish-looking integrand. I'll try to slug it out and see what develops though. $\endgroup$ – J. M. is a poor mathematician Aug 17 '10 at 1:40
  • 1
    $\begingroup$ Mathematica may be introducing the hypergeometric function to deal with the integral of $\cos^{n}\psi$ symbolically in $n$. You can avoid this by invoking the reduction formula $I(\cos^n\psi)=\frac{n-1}{n} I(\cos^{n-2}\psi)$, where $I()$ is the integral from 0 to 2pi. (You won't even need the Gamma function to express the end result.) Other than that, the terms of your integrand involve factors of the form $\cos^{even} \sin^{odd}$ in $\theta$ and $\phi$; expressing $\sin^{odd}$ as $(1-\cos^2)^m \sin$ sets you up for a $u=\cos$ substitution. Making sense of the binomial soup is up to you. :) $\endgroup$ – Blue Aug 18 '10 at 1:09
  • $\begingroup$ Re my claim about odds and evens: The terms of your integrand have the form $\cos^j \theta \sin^{2k-j+1}\theta \cos^j\phi \sin^{2k-j+1}\phi \cos^{2k-j}\psi$. Since you're integrating $\psi$ across the full period of $\cos$, terms with odd powers of $\cos\psi$ vanish. Therefore, you may take $j$ to be even, giving even powers of cosine and odd powers of sine. $\endgroup$ – Blue Aug 18 '10 at 1:14
  • $\begingroup$ Probably the galling thing about it is that the hypergeometric function keeps on cropping up even after the use of Assuming[] to tell Mathematica that I have integer variables; on the other hand, I also noticed during my preliminary analysis that the binomial sum from 0 to 2k can be converted into a sum ranging from 0 to k due to odd powers of cosine vanishing when integrated from 0 to $2\pi$. I'll look into your approach and report back. Thanks! $\endgroup$ – J. M. is a poor mathematician Aug 18 '10 at 1:55

Okay ... here we go ...

To save typing, I'm going to use $A$, $B$, and $C$ for $\theta$, $\phi$, and $\psi$. So, our integral starts as this ...

$$ \int\int\int \left(2 \cos A \cos B - \sin A \sin B \cos C\right)^{2k} \sin A \sin B $$

... and expands to this ...

$$\int\int\int \sum_{j=0}^{2k} {2k \choose j} \; 2^j (-1)^{2k-j} \cos^j A \; \sin^{2k-j+1} A \; \cos^j B \; \sin^{2k-j+1} B \; \cos^{2k-j} C$$

The integral over the sum will be a sum of integrals over the terms; and, since the $A$, $B$, and $C$ factors are independent, each integrated term is a product of integrals. Moreover, as mentioned in my comment on the original question, we're integrating $\cos^{2k-j}C$ over the full period of cosine, so terms with odd powers of $\cos C$ --those corresponding to odd values of $j$-- vanish. So, we take $j=2h$ and for convenience define $m := k-h$, then we write ...

$$\sum_{h=0}^{k} 2^{2h} {2k \choose 2h} \; \left( \int \cos^{2h} A \; \sin^{2m+1} A \right) \left( \int \cos^{2h} B \; \sin^{2m+1} B \right) \left( \int \cos^{2m} C \right)$$

The first two integrals are identical, so we're really looking at the following:

$$\sum_{h=0}^{k} 2^{2h} {2k \choose 2h} \; \left( \int \cos^{2h} A \; \sin^{2m+1} A \right)^2 \left( \int \cos^{2m} C \right)$$

Consider the pieces ...

By the reduction formula for the integral of powers of cosine, we have

$$\int_{0}^{2\pi} \cos^{2m}C \; dC = \frac{(2m-1)(2m-3)\cdots(1)}{(2m)(2m-2)\cdots(2)}\int_{0}^{2\pi} dC = \frac{2\pi(2m-1)!!}{2^{m} m!} = \frac{2\pi(2m)!}{2^{2m} m!^2}$$

where $x!!$ is the double factorial.

As for the other factor ...

$$\int_{0}^{\pi} \cos^{2h} A \; \sin^{2m+1} A \; dA = \int_{0}^{\pi} \cos^{2h} A \; \left( 1 - \cos^{2}A\right)^m \; \sin A \; dA = \int_{-1}^{1} u^{2h} \left( 1 - u^2 \right)^m \; du$$

$$= \int_{-1}^{1} u^{2h} \sum_{n=0}^{m} {m \choose n} (-1)^n u^{2n} \; du = \sum_{n=0}^{m} (-1)^n {m \choose n} \int_{-1}^{1} u^{2n+2h} \; du = \sum_{n=0}^{m} {m \choose n} \frac{2 (-1)^n}{2n+2h+1} $$

Mathematica evaluates the sum into a form we manipulate into something we can use:

$$\frac{\Gamma(h+\frac{1}{2}) m!}{\Gamma(m+h+\frac{3}{2})} = \frac{\sqrt{\pi} (2h-1)!!}{2^h}\frac{2^{k+1}m!}{\sqrt{\pi}(2k+1)!!}= \frac{(2h)!}{2^{2h}h!}\frac{2^{k+1}m!}{(2k+1)}\frac{2^{k}k!}{(2k)!}= \frac{2^{2m+1}m!}{2k+1}\frac{(2h)!}{h!}\frac{k!}{(2k)!}$$

This leads to a simplification bonanza in the formula for the integral ...

$$\sum_{h=0}^{k} \;\; 2^{2h} \;\;\;\; \frac{(2k)!}{(2h)!(2m)!} \;\;\;\; \frac{2^{4m+2}m!^2}{(2k+1)^2}\frac{(2h)!^2}{h!^2}\frac{k!^2}{(2k)!^2} \;\;\;\; \frac{2\pi(2m)!}{2^{2m} m!^2}$$

$$= 8 \pi \sum_{h=0}^{k} \frac{2^{2(m+h)}}{(2k+1)^2}\frac{(2h)!}{h!^2}\frac{k!^2}{(2k)!}$$

$$= 8 \pi \frac{2^{2k}}{(2k+1)^2} \frac{k!^2}{(2k)!} \sum_{h=0}^{k}\frac{(2h)!}{h!^2}$$

$$= 8 \pi \left(\frac{2^{k}}{2k+1}\right)^2 {2k \choose k}^{-1} \;\; \sum_{h=0}^{k}{2h \choose h}$$

... as expected (well, "expected" from your having previously forced it out of Mathematica).

  • 2
    $\begingroup$ You sir, may be a day late, but a prince otherwise. :) Thank you! I suppose I owe you and other readers an explanation on the provenance of this triple integral so here goes: this appears as a coefficient in the series expansion of the "second virial coefficient" for the equation of state of a polar fluid. The integration over three angles represents the three degrees of freedom a polar molecule can move in a fluid. The triple integral appears in the book "Molecular Theory of Gases and Liquids" by Curtiss et al., but there is not even a hint that a simpler form for this exists. $\endgroup$ – J. M. is a poor mathematician Aug 18 '10 at 11:23
  • 2
    $\begingroup$ Glad to be of service. Of course, knowing what the answer was supposed to be helped a great deal. :) Interestingly, if you steal a factor of 2 from the 8, you can re-write the fraction as $\frac{2^{2k+1}}{(2k+1)^2}$, where the numerator and denominator swap bases and exponents. (And, for some reason, $4\pi$ seems like a more natural multiplier than $8\pi$ anyway. :) $\endgroup$ – Blue Aug 18 '10 at 11:59

Here is a hint to an other approach: Search for the key words "convolution formula" "Legendre polynomial" "addition formula" and you will be en-lighted...

--- A little explanation and some spherical harmonics ---

@J. M. You are totally right in particular when I was wrong in my suggestion! :/

What I thought about was convolution on the double coset space $X=K\backslash G/K$ of $G= SO(3)$ where $K=SO(2)$. The space $X$ may be identified with $[-1,1]$ and the convolution (as induced from the translation operator on $G$) is then given by $$f*g(x)= \frac{1}{4\pi}\int_{-1}^1\int_{0}^{2\pi}f(y)g(xy+\sqrt{1-x^2}\sqrt{1-y^2}\cos\theta) d\theta dy.$$ An interesting fact is that $f*g=g*f$ even though $G$ is not Abelian. Choosing $f(x)=1$ and $g(x)=x^{2k}$ we get $$\frac{1}{8\pi}\int_{-1}^1\int_{-1}^1\int_{0}^{2\pi}(xy+\sqrt{1-x^2}\sqrt{1-y^2}\cos\theta)^{2k} d\theta dydx=$$ $$\int_{-1}^1f*g(x)dx/2 = \int_{-1}^1 g*f(x)dx/2=\int_{-1}^1x^{2k}dx/2= 1/(2k+1).$$ In our case the integrand is (after changing variables) $$(2xy + \sqrt{1-x²}\sqrt{1-y²}\cos\theta)^{2k}=$$ $$=\sum_{j=0}^{2k}{2k \choose j}(xy)^j(xy+\sqrt{1-x²}\sqrt(1-y²)\cos\theta)^{2k-j}$$ Using this we get an other convolution - but I can not see how this can be used directly. :(

Legendre polynomials might be good since they are the characters on the space $X$.

  • $\begingroup$ Some elaboration would be welcome. :) $\endgroup$ – J. M. is a poor mathematician Aug 18 '10 at 11:23
  • $\begingroup$ @J. Mangaldan: I changed the answer a bit.. sorry :) $\endgroup$ – AD. Aug 21 '10 at 8:31
  • $\begingroup$ I would probably have been much more interested if you showed it as something involving Chebyshev polynomials; for Legendre polynomials on the other hand, I could not remember any property of them suitable for the simplification of trigonometric integrals, which is why I was skeptical. $\endgroup$ – J. M. is a poor mathematician Aug 21 '10 at 8:40

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.