1

I would like to solve the following Langevin equation $$\frac{d^2 x}{d t^2}+\omega_0^2x(t)=\eta(t),$$ where $\eta(t)$ is a blue noise signal given by $$\eta(t)=\int_{-\infty}^\infty \hat{\eta}(f)\exp(2\pi i t f) df,$$ and $\hat{\eta}(t)$ is the fourier transform given by $$\hat{\eta}(f)=\int_{-\infty}^\infty \eta(t)\exp(-2\pi i t f) dt,$$ such that $|\hat{\eta}(f)|=D\sqrt{|f|}$, where $D$ is a constant which gives a measure of the strength of the signal. According to Wikipedia the autocorrelation is given by $$\begin{aligned} R_{\eta \eta}(\tau)&=\int_{-\infty}^\infty|\hat{\eta}(f)|^2\exp(2\pi i \tau f) df,\\ &= \frac{D^2}{2\pi^2\tau^2}, \end{aligned}$$ where the Fourier transform is calculated using an identity in Wikipedia. This post shows that the variance is given by $$\begin{aligned} \langle x^2(t) \rangle &= \frac{1}{\omega_0^2}\int_0^t\int_0^t \sin[\omega_0(t-t')]\sin[\omega_0(t-t'')]R_{\eta\eta}(t'-t'')dt''dt' \\ &= \frac{D^2}{2\pi\omega_0^2}\int_0^t\int_0^t \frac{\sin[\omega_0(t-t')]\sin[\omega_0(t-t'')]}{(t'-t'')^2}dt''dt'. \end{aligned}$$ This integral is not defined at $t'=t''$, does this mean we cannot calculate the solution for the case where $\eta(t)$ is blue noise or have I made a mistake in the maths?

Peanutlex
  • 1,169
  • So the auto correlation $R$ at $\tau=0$ is $-\infty$? – user619894 Jul 22 '20 at 12:02
  • Yes but that makes sense right? The autocorrelation at $\tau=0$ gives the variance. The power spectrum is given by $S_{\eta\eta}(f)=D^2|f|$, so all frequencies have non-zero energy associated with them. Therefore we expect the energy or the variance of $\eta(t)$ to be infinite. The white noise driver in the linked post is also infinite for $\tau=0$. – Peanutlex Jul 22 '20 at 12:11
  • I think it should be $+\infty$ and I have now edited the main post. – Peanutlex Jul 22 '20 at 12:17
  • 1
    If you are trying to solve an equation with infinite noise, you shouldn't be surprised you get strange results. What happens if you regularize the noise, say work with $R(\tau)\sim {1\over \tau_{0}^{2}+\tau^2} $? – user619894 Jul 22 '20 at 13:02
  • I guess I am hopeful that given white noise works okay, maybe blue noise works as well. – Peanutlex Jul 22 '20 at 13:53
  • Also, can you point out which formula you used to calculate $R$, I couldn't find it in the table you linked to . – user619894 Jul 22 '20 at 13:56
  • I used 311 in that table. – Peanutlex Jul 22 '20 at 14:06
  • 1
    Not valid for $\alpha=1$ – user619894 Jul 22 '20 at 14:19
  • You are right, however, we can derive it using $|x|=x,\text{sgn}(x)$, then using 109, 308 and 312. – Peanutlex Jul 22 '20 at 14:27
  • 1
    still think a $\delta$ function is missing. Consider the $FT$ of $\alpha\rightarrow 0$ limit of ${\exp^{-\alpha |x|}\over \alpha}$ – user619894 Jul 22 '20 at 14:47
  • Thank you that is a really good idea. I guess you mean $[1-\exp(-\alpha |x|)]/\alpha$? That does give a delta function. – Peanutlex Jul 22 '20 at 14:57
  • 1
    Yes, I picked the one without, to show you that its FT goes to $1/w^2$, so that $1/w^2$ cannot represent $|x|$ without some extra $\delta$ – user619894 Jul 22 '20 at 15:11
  • Do you know if there is a simpler form for the Fourier transform of $|x|$ because that would make the integral easier? – Peanutlex Jul 22 '20 at 15:38

0 Answers0