I have a Fredholm integral equation of second kind

$$\frac{-1}{2\pi \omega'^2}+\int_{-\infty}^{\infty}\frac{1}{\pi \omega'^2}\frac{\psi(\omega)d\omega}{{\text{sinch}}((\omega-\omega')\pi/2)}=\psi(\omega'),$$

where $\text{sinch}(x)=\sinh(x)/x.$ I've been trying to solve for $\psi$ by putting $\psi(\omega)=\sum_{n=0}^\infty a_n\omega^n$. However the integral $$a_n\int_{-\infty}^{\infty}\frac{1}{\pi \omega'^2}\frac{\omega^nd\omega}{{\text{sinch}}((\omega-\omega')\pi/2)},$$ seems to be cumbersome.

Any thoughts on how to solve these types of integrals would be appreciated. I also failed to find it in any integral table.



It is sufficient to compute $$ \int_{-\infty}^\infty \frac{w^n dw}{\mathrm{sinch}(\pi w/2)}, $$ since you can shift by $w'$ and expand the numerator in the original integral. Now, this integral converges very well, and there is no harm in adding a factor $\exp(i\epsilon\omega)$ to the integrand, for any $\epsilon>0$. However, now you can deform the contour into the upper half-plane with no harm for the convergence. In this way you can reduce to sum of the residues at the poles $w=2ik$, $k=1,2,\ldots$ $$ \int_{-\infty}^\infty \frac{w^n e^{i\epsilon w} dw}{\mathrm{sinch}(\pi w/2)}=2\pi i\sum_{k=1}^\infty \mathrm{res}(2ik)=-4\pi \sum_{k=1}^\infty(2ik)^nk(-1)^ke^{-2\epsilon k}=2\pi i^n(-\partial_\epsilon)^{n+1}\frac{1}{e^{2\epsilon}+1}. $$ Now, we have for Euler polynomials $$ \frac{2e^{xt}}{e^t+1}=\sum_{k=0}^\infty E_k(x)\frac{t^k}{k!}, $$ so that $$ \frac{1}{e^{2\epsilon}+1}=\sum_{k=0}^\infty 2^{k-1}E_k(0)\frac{\epsilon^k}{k!}, $$ giving for the integral $$ \left.\int_{-\infty}^\infty \frac{w^n e^{i\epsilon w} dw}{\mathrm{sinch}(\pi w/2)}\right\vert_{\epsilon=0}=\left.2\pi i^n(-\partial_\epsilon)^{n+1}\frac{1}{e^{2\epsilon}+1}\right\vert_{\epsilon=0}=-2\pi (-2i)^nE_{n+1}(0). $$ So finally $$ \int_{-\infty}^\infty \frac{w^n dw}{\mathrm{sinch}(\pi w/2)}=-2\pi (-2i)^nE_{n+1}(0). $$ This is always real, since both sides vanish for odd $n$.

Note that the Euler polynomials form an Appell sequence, and so have a fundamental property $$ E_n(x+y)=\sum_{k=0}^n \binom{n}{k}E_k(x)y^{n-k}, $$ so a better statement of the result is that (note the change from $\mathrm{sinch}$ to $\sinh$) $$ \int_{-\infty}^\infty \frac{(w+w')^n dw}{\mathrm{sinh}(\pi w/2)+i0}=(-2i)^{n+1}E_{n}\left(\frac{iw'}{2}\right), $$ where $+i0$ in the denominator tells you that your integration contour should go above the pole at $w=0$ in the complex plane. Curiously, this integral representation of Euler polynomials seems to be absent from Wolfram functions site. However, I guess it should be known.

  • 2
    $\begingroup$ To meet the formula of the OP the nominator of the integrand in the very last formula should be $w (w+w')^{n-1}$, which then gives a less neat result. If you want to (in a sense) double check: The very first integral in the answer can also be found in Gradshteyn & Ryshik, formula 3.523 2. $\endgroup$ May 9 '15 at 10:57
  • 2
    $\begingroup$ @JohannesTrost, Thanks for the reference! Regarding the integral of the OP, it can be easily reduced to the integral computed in the middle of the answer. I just included the neat version because it is neat :) $\endgroup$ May 9 '15 at 23:12

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.