2

A buddy and I are hung up on this integral.

Prove that:

$\displaystyle \int_{0}^{\infty}e^{-ax^{2}} \sin(b/x^{2})dx=-\text{Im}\int_{0}^{\infty}e^{-ax^{2}-ib/x^{2}}dx=\frac{1}{2}\sqrt{\frac{\pi}{a}}\frac{\sin(\sqrt{2ab})}{e^{\sqrt{2ab}}}$

$\displaystyle\int_{0}^{\infty}e^{-ax^{2}-b/x^{2}}dx=\frac{1}{2}\sqrt{\frac{\pi}{a}}\frac{1}{e^{\sqrt{2ab}}}$ was used.

The thing is, this goes under the assumption that $b$ is imaginary. Isn't this quite a leap to make without justification?

Does anyone know of a good way to prove this?

Cody
  • 14,054
  • 1
    It is a bit hard to guess what precisely you are asking. Could you please make this clearer? – Rasmus Jan 05 '12 at 12:44
  • I'm sorry. Prove that $\displaystyle\int_{0}^{\infty}e^{-ax^{2}}sin(b/x^{2})dx=\frac{1}{2}\sqrt{\frac{\pi}{a}}\frac{sin(\sqrt{2ab})}{e^{\sqrt{2ab}}}$ – Cody Jan 05 '12 at 13:50
  • 1
    For which values of $a$ and $b$? – Rasmus Jan 05 '12 at 14:50
  • @Cody have you tried the silver bullet for integration i.e. wolphram alpha ? :) – nb1 Jan 05 '12 at 14:52
  • 1
    It said "prove", so perhaps "Wolfram Alpha" does not qualify. – GEdgar Jan 05 '12 at 15:43
  • Maybe the answer is that (at least the real part of) the integral is just zero for real $b$. If you write $b=b_1+ib_2$, then you get your second integral, where additionally the integrand is multiplied by $e^{i\frac{-b_1}{x^2}}$. My guess is that for small $x$, this oscillation $e^{i\infty}$ kills the integral and for large $x$ the integrand falls of due to the gauss curve anyway. – Nikolaj-K Jan 05 '12 at 16:13

1 Answers1

1

Consider the formula $$ \int_{0}^{\infty}e^{-ax^{2}-b/x^{2}}dx=\frac{1}{2}\sqrt{\frac{\pi}{a}}\frac{1}{e^{\sqrt{2ab}}} $$ In the domain of pairs $a,b$ of complex numbers where it converges, the left-hand side is an analytic function of $a$ and $b$. There is a sensible domain where the right-hand side is an analytic function of $a$ and $b$. Therefore, if the two are equal on a large enough set (for example, all positive real $a,b$), then it is true on the entire domain of convergence. So, we only need to check that there is a sensible domain for this that includes imaginary $b$.

GEdgar
  • 111,679