While studying the use of the trapezoidal rule for numerically evaluating the complementary error function $\mathrm{erfc}(z)$, the following integrals showed up when I was trying to derive expressions for the truncation error:

$$\int_0^\pi \exp\left(-z^2\tan^2\frac{u}{2}\right)\cos(2mu) \mathrm du$$

where $z$ is positive and $m$ is a positive integer.

Evaluating a bunch of these integrals in Mathematica, I gather that these integrals follow the pattern

$$\pi z^2\exp(z^2)\mathrm{erfc}(z)R_n(z)-2\sqrt{\pi}z S_n(z)$$

where $R_n(z)$ and $S_n(z)$ are polynomials.

Are there any closed forms for these two polynomials?


I played around with this for a while, but didn't find a complete answer. Anyway, here's some partial information, in case that's of any help.

The substitution $t=\tan(u/2)$ turns your integral into $$\int_0^\infty e^{-z^2 t^2} \cos(4 m \arctan t) \frac{2 dt}{1+t^2}.$$ Since $\exp(i \arctan t) = \frac{1+it}{\sqrt{1+t^2}}$, this can be written as $$\Re \int_{-\infty}^\infty e^{-z^2 t^2} \frac{(1+it)^{4m}}{(1+t^2)^{2m+1}} dt = \sum_{k=0}^{2m} \int_{-\infty}^\infty e^{-z^2 t^2} \frac{\binom{4m}{2k}(-1)^k t^{2k}}{(1+t^2)^{2m+1}} dt.$$ By writing $t^{2k} = ((t^2+1)-1)^k$ and expanding using the binomial theorem, this integral can be reduced to a sum of integrals of the form $$ J_n(z) = \int_{-\infty}^\infty e^{-z^2 t^2} \frac{1}{(1+t^2)^n} dt.$$ We have $J_0(z) = \sqrt{\pi}/z$ right away, and also $J_1(z) = \pi \exp(z^2) \mathrm{erfc}(z)$; the latter equality follows since both sides satisfy the differential equation $(d/dz)(\exp(-z^2) f(z)) = -2 \sqrt{\pi} \exp(-z^2)$, and both sides agree at $z=0$. Moreover, the following two-term recursion relation holds: $$ (2m-2) J_m(z) - (2m-3-2z^2) J_{m-1}(z) - 2z^2 J_{m-2}(z) = 0.$$ Proof: Verify by direct calculation that left-hand side is the integral of $\frac{\partial}{\partial t} \left( \frac{t \exp(-t^2 z^2)}{(1+t^2)^{(m-1)}} \right)$.

From this it follows that each $J_n$ is a linear combination of $J_0$ and $J_1$ with coefficients that are polynomials in $z^2$, and this also implies that the pattern that you've observed really is correct, but it seems tricky to get a nice explicit form for those polynomials...


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.