Is there a way to show that

$$\int_{-\infty}^{\infty} \int_{-\infty}^{\infty} (\tau \xi - \sigma \eta) \frac{\exp(\sigma\xi + \tau\eta + c)}{[1 + \exp(\sigma\xi + \tau\eta + c)]^2} \exp(\frac{-\xi^2}{2}) \exp(\frac{-\eta^2}{2}) d\xi d\eta = 0$$

where $\sigma, \tau > 0$ and $c \in \mathbb{R}$?

Numerically, with special cases, I can see it's true. I wonder if there's some sleek trick. You can even see that the integral is probably 0 due to the symmetry in the plot of the integrand, where z4 is the area and $x = \xi, y = \eta$:

Plot of the integrand when $\sigma = 1, \tau = 2, c = 7$

Thus, I wonder if the trick is to split up the positive parts (when $\tau \xi > \sigma \eta$) and the negative parts (when $\tau \xi < \sigma \eta$), and show that $$\int_{-\infty}^{\infty} \int_{-\infty}^{\frac{\sigma \eta}{\tau}} (\tau \xi + \sigma \eta) \frac{\exp(\sigma\xi + \tau\eta + c)}{[1 + \exp(\sigma\xi + \tau\eta + c)]^2} \exp(\frac{-\xi^2}{2}) \exp(\frac{-\eta^2}{2}) d\xi d\eta \ + \\ \int_{-\infty}^{\infty} \int_{\frac{\sigma \eta}{\tau}}^{\infty} (\tau \xi + \sigma \eta) \frac{\exp(\sigma\xi + \tau\eta + c)}{[1 + \exp(\sigma\xi + \tau\eta + c)]^2} \exp(\frac{-\xi^2}{2}) \exp(\frac{-\eta^2}{2}) d\xi d\eta = 0, $$ where the first part on the left-hand side is the positive part and the second part on the left-hand side is the negative part.

So we must show: $$\int_{-\infty}^{\infty} \int_{-\infty}^{\frac{\sigma \eta}{\tau}} (\tau \xi + \sigma \eta) \frac{\exp(\sigma\xi + \tau\eta + c)}{[1 + \exp(\sigma\xi + \tau\eta + c)]^2} \exp(\frac{-\xi^2}{2}) \exp(\frac{-\eta^2}{2}) d\xi d\eta \ = \\ -\int_{-\infty}^{\infty} \int_{\frac{\sigma \eta}{\tau}}^{\infty} (\tau \xi + \sigma \eta) \frac{\exp(\sigma\xi + \tau\eta + c)}{[1 + \exp(\sigma\xi + \tau\eta + c)]^2} \exp(\frac{-\xi^2}{2}) \exp(\frac{-\eta^2}{2}) d\xi d\eta$$

Any ideas? This is definitely not a HW question. It's something I came across while working out a derivation.

  • 1
    $\begingroup$ The change of variable $$(x,y)=(\tau\xi-\sigma\eta,\sigma\xi+\tau\eta)$$ allows to write this double integral as the product of two simple integrals, one of whose is zero by symmetry. $\endgroup$ – Did Dec 16 '17 at 10:05
  • $\begingroup$ @Did : How did you get a product of two simple integrals? $\endgroup$ – 193381 Dec 16 '17 at 20:58
  • $\begingroup$ @Did : After your change of variable, I need to show this: $$\frac{1}{\tau^2 + \sigma^2} \int_{-\infty}^{\infty} \int_{-\infty}^{0} x \frac{\exp(v + c)}{[1 + \exp(v + c)]^2} \exp\left(\frac{- (\tau u + \sigma v) ^2}{2(\tau^2 + \sigma^2)^2}\right) \exp\left(\frac{- (\tau u - \sigma v) ^2}{2(\tau^2 + \sigma^2)^2}\right) dx dy \ = \\ \frac{1}{\tau^2 + \sigma^2} \int_{-\infty}^{\infty} \int_{0}^{\infty} x \frac{\exp(v + c)}{[1 + \exp(v + c)]^2} \exp\left(\frac{- (\tau u + \sigma v) ^2}{2(\tau^2 + \sigma^2)^2}\right) \exp\left(\frac{- (\tau u - \sigma v) ^2}{2(\tau^2 + \sigma^2)^2}\right) dx dy $$ $\endgroup$ – 193381 Dec 16 '17 at 21:18
  • $\begingroup$ @Did : I think the $c$ is the issue... fuks things up $\endgroup$ – 193381 Dec 16 '17 at 21:18
  • 1
    $\begingroup$ Simply develop and sum the exponents in the exponentials... to see what you think is an "issue" disappear. $\endgroup$ – Did Dec 16 '17 at 21:20

Write $\mathrm{a} = (\tau, \sigma)$ and $\mathrm{x} = (\eta, \xi)$. Also let $R_{\theta}$ be the rotation by $\theta$-radian and $J = R_{\pi/2}$. Then your integral can be reformulated as

$$ \int_{\mathbb{R}^2} \frac{4\langle J\mathrm{a}, \mathrm{x} \rangle}{\cosh^2(\frac{1}{2}\langle \mathrm{a}, \mathrm{x} \rangle + \frac{c}{2})} e^{-|\mathrm{x}|^2} \, \mathrm{d}\mathrm{x}$$

Recognizing this integral as function of $\mathrm{a}$ and denoting this by $I(\mathrm{a})$, we find that for any $\theta \in \mathbb{R}$,

\begin{align*} I(\mathrm{a}) &= \int_{\mathbb{R}^2} \frac{4\langle JR_{\theta}\mathrm{a}, R_{\theta}\mathrm{x} \rangle}{\cosh^2(\frac{1}{2}\langle R_{\theta} \mathrm{a}, R_{\theta}\mathrm{x} \rangle + \frac{c}{2})} e^{-|R_{\theta}\mathrm{x}|^2} \, \mathrm{d}\mathrm{x} \\ &= \int_{\mathbb{R}^2} \frac{4\langle JR_{\theta}\mathrm{a}, \mathrm{x}' \rangle}{\cosh^2(\frac{1}{2}\langle R_{\theta} \mathrm{a}, \mathrm{x}' \rangle + \frac{c}{2})} e^{-|\mathrm{x}'|^2} \, \mathrm{d}\mathrm{x}' \qquad (\mathrm{x}' = R_{\theta}\mathrm{x})\\ &= I(R_{\theta}\mathrm{a}) \end{align*}

In the first line, we utilized the fact that rotations are isometries, i.e., $\langle \mathrm{x}, \mathrm{y} \rangle = \langle R_{\theta}\mathrm{x}, R_{\theta}\mathrm{y} \rangle$ always holds, and the fact that any two 2D rotations commute, i.e. $R_{\theta}R_{\psi} = R_{\psi}R_{\theta}$.

So by applying a rotation, we may assume that $\mathrm{a} = (a, 0)$, where $a = \sqrt{\tau^2+\sigma^2}$. Then

$$ I(a,0) = \int_{\mathbb{R}^2} \frac{4 a \xi }{\cosh^2(\frac{1}{2}a\eta + \frac{c}{2})} e^{-\eta^2-\xi^2} \, \mathrm{d}\eta\mathrm{d}\xi = 0 $$

since the integrand is odd with respect to $\xi$.

  • $\begingroup$ ah, rotation. I need to brush up on that. Thank you for answering. $\endgroup$ – 193381 Dec 16 '17 at 21:21

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.