I've been examining an integral I remember encountering in complex analysis:

$$ I = \int_0^\pi e^{\cos\theta} \cos\left(n\theta - \sin\theta\right) d\theta $$

where $n$ is a nonnegative integer. It's not too hard to show that $I = \frac{\pi}{n!}$ by changing $\cos\left(n\theta - \sin\theta\right)$ into $\operatorname{Re}\left[e^{i(n\theta - \sin\theta)}\right]$ and making the substitution $z = e^{i\theta}$ to turn the integral into a contour integral on the complex unit circle.

However, I came across the following representation of the Bessel functions of the first kind on Wikipedia:

$$ J_n(x) = \frac{1}{\pi} \int_0^\pi \cos\left(n\theta - x\sin\theta\right) d\theta $$

Given the similarity between this formula when $x = 1$ and the initial integral, I'm sure there's some way to evaluate the integral using properties of the Bessel functions. But so far I'm stuck as to figuring it out. I tried expanding $e^{\cos\theta}$ as a series but that got ugly pretty quickly. Any suggestions?


If we compute the Fourier cosine series of $e^{\cos\theta}$ over $[-\pi,\pi]$ we get: $$ e^{\cos\theta} = I_0(1)+2\sum_{k\geq 1}I_k(1)\cos(k\theta)\tag{1}$$ where $I_k$ is a modified Bessel function of the first kind. On the other hand, by using Briggs formulas and your definition of $J_n(x)$ it follows that: $$ \frac{1}{2\pi}\int_{-\pi}^{\pi}\cos(n\theta-\sin\theta)\,d\theta = J_n(1),$$ $$ \frac{1}{\pi}\int_{-\pi}^{\pi}\cos(n\theta-\sin\theta)\cos(k\theta)\,d\theta = J_{n+k}(1)+J_{n-k}(1),\tag{2}$$ so we have: $$ \int_{-\pi}^{\pi}e^{\cos\theta}\cos(n\theta-\sin\theta)\,d\theta =\\2\pi \cdot I_0(1)\,J_n(1)+2\pi\cdot\sum_{k\geq 1}I_k(1)\left(J_{n-k}(1)+J_{n+k}(1)\right)\tag{3}$$ and now we may exploit: $$ \sum_{n=-\infty}^{+\infty}t^n\,J_n(z) = \exp\left(\frac{z}{2}\left(t-\frac{1}{t}\right)\right) \tag{4}$$ together with $I_n(z)=(-i)^n\,J_n(iz)$ to write $(3)$ as a convolution and deduce: $$ I = \pi\cdot [z^n]\,e^{z} = \frac{\pi}{n!}\tag{5} $$ in a extremely convoluted and probably useless very nice way.

  • $\begingroup$ Thanks! I'm still a little unclear on the convolution part. I understand up to where $I = \pi \sum_{k=-\infty}^\infty I_k(1) J_{n-k}(1) = \pi \sum_{k=-\infty}^\infty (-i)^k J_k(i) J_{n-k}(1)$ but I don't see how to apply eqn (4) to that to get eqn (5). $\endgroup$ – Billy R Jun 1 '15 at 17:42
  • 1
    $\begingroup$ @BillyR: just read the RHS of $(3)$ as the coefficient of $t^n$... in the right product. $\endgroup$ – Jack D'Aurizio Jun 1 '15 at 18:11
  • $\begingroup$ Ahh, I see it now. Thanks again! $\endgroup$ – Billy R Jun 1 '15 at 19:11

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.