Consider a stochastic differential equation: $$ dx(t) = a x(t)dt + b y(t)dt \quad (1) $$ where $y(t)$ is a stochastic process satisfying $\langle y(t)y(t')\rangle = \delta(t-t')$. We will assume that the first statistical moments are zero and that $a<0$. I am interested in the variance $\langle x(t)^2 \rangle$. When I solve this equation in the frequency domain, I obtain the value $\langle x(t)^2 \rangle = \frac{-b^2}{2a}$. This is the steady state variance,($\lim_{x\rightarrow \infty} \langle x(t)^2 \rangle$) but does not give any information about the transient variance $\langle x(t)^2 \rangle$ at $t\neq \infty$.

My question is: why does solving this equation in the frequency domain not capture the transient behavior of the variance?

More mathematical detail are below:

My question is: why does solving in the frequency domain give the steady state variance, and not a function which depends on $t$?

We can be solve in the time domain to give: $$ \langle x(t)^2 \rangle = e^{2at}\langle x(0)^2 \rangle - \frac{b^2}{2a}(1-e^{2at}) $$ The steady state can then be found by taking $t \rightarrow \infty$.

We can also solve in the frequency domain. Fourier transforming $(1)$ gives: $$ x(\omega) = \frac{b}{-a-i\omega} y(\omega) \quad (2) $$ where $y(\omega)$ now obeys $\langle y(\omega)y(\omega')\rangle = \delta(\omega + \omega')$. We then invert the Fourier transform to find the variance:

$$ \langle x(t)^2 \rangle = \frac{1}{2 \pi} \int^{\infty}_{-\infty}\int^{\infty}_{-\infty}\langle x(\omega)x(\omega')\rangle^{-it(\omega + \omega')} d\omega d\omega' \quad (3) $$ Plugging $(2)$ into $(3)$ and integrating over the delta function gives a value of $\frac{-b^2}{2a}$, which is the steady state variance obtained earlier, but does not contain any information about the transient behavior. Why is this?

  • $\begingroup$ Can you show your time-domain analysis? When Fourier transforming the original differential equation, I dealt with $e^{-i\omega t}dx(t)$ by integrating $d(e^{-i\omega t}x(t)) + i\omega e^{-i\omega}x(t)dt$ and assuming that the boundary terms arising from integrating the first term, are zero. Is that an error? Perhaps $\hat{x}(\omega) = b\hat{y}(\omega)/(i\omega - a) + c$ for some nonzero constant $c$. $\endgroup$ – Joe Mack Jun 18 at 17:53

BOTTOM LINE UP FRONT: I think the exponential decay growth in $\left<|x(t)|^2\right>$ can be shown in the frequency domain only if the "boundary terms" are nonzero when we compute the Fourier transform of $dx(t)$ from the original SDE.

I provide only a start in the work below.

Since these processes seem like they could possibly be complex-valued, I will consider $\left<x(t)\overline{x(t')}\right>$ and $\left<y(t)\overline{y(t')}\right>$, where the $\texttt{overline}$ indicates complex conjugation.
I will write out some steps, because I need to see some things that you did not write explicitly.
\begin{equation} dx(t) = ax(t)dt + by(t)dt. \end{equation} Fourier transform: \begin{equation} \begin{split} \int e^{-i\omega t}dx(t) &=~ \int e^{-i\omega t}ax(t)dt + \int e^{-i\omega t} by(t)dt\\ \int\left[d\left(e^{-i\omega t}x(t)\right) - x(t)d(e^{-i\omega t})\right] &=~ a\widehat{x}(\omega) + b\widehat{y}(\omega)\\ \underbrace{\int d\left(e^{-i\omega t}x(t)\right)}_{\textrm{Boundary terms: $-c$, a constant}} - \int x(t)(-i\omega)e^{-i\omega t}dt &=~ a\widehat{x}(\omega) + b\widehat{y}(\omega)\\ i\omega\widehat{x}(\omega) &=~ a\widehat{x}(\omega) + b\widehat{y}(\omega) + c \end{split} \end{equation} Result: \begin{equation} \widehat{x}(\omega) = \frac{b\widehat{y}(\omega) + c}{i\omega - a} \end{equation}
\begin{equation} \begin{split} x(t) &=~ \frac{1}{2\pi}\int \widehat{x}(\omega) e^{it\omega}d\omega\\ &=~ \frac{1}{2\pi}\int \frac{b\widehat{y}(\omega) + c}{i\omega - a} e^{it\omega}d\omega\\ & \\ x(t') &=~ \frac{1}{2\pi}\int \widehat{x}(\omega') e^{it'\omega'}d\omega'\\ &=~ \frac{1}{2\pi}\int \frac{b\widehat{y}(\omega') + c}{i\omega' - a} e^{it'\omega'}d\omega' \end{split} \end{equation} \begin{equation} \end{equation} \begin{equation} \begin{split} \left<x(t)\overline{x(t')}\right> &=~ \left<\frac{1}{2\pi}\int \frac{b\widehat{y}(\omega) + c}{i\omega - a} e^{it\omega}d\omega\frac{1}{2\pi}\int \frac{\overline{b\widehat{y}(\omega') + c}}{-i\omega' - a} e^{-it'\omega'}d\omega'\right>\\ &=~ \frac{1}{4\pi^2}\int\int\frac{b^2\left<\widehat{y}(\omega)\overline{\widehat{y}(\omega')}\right> + bc\left<\overline{\widehat{y}(\omega')}\right> + b\overline{c}\left<\widehat{y}(\omega)\right> + |c|^2}{(i\omega - a)(-i\omega' - a)}e^{it\omega}e^{-it'\omega'}d\omega d\omega' \end{split} \end{equation}
\begin{equation} \begin{split} \left<\widehat{y}(\omega)\overline{\widehat{y}(\omega')}\right> &=~ \left<\int y(\tau)e^{-i\omega\tau}d\tau\overline{\int y(\tau')e^{-i\omega'\tau'}d\tau'}\right>\\ &=~ \int\int\left<y(\tau)\overline{y(\tau')}\right>e^{-i\omega\tau}e^{i\omega'\tau'}d\tau d\tau'\\ &=~ \int\int\delta(\tau-\tau')e^{-i\omega\tau}e^{i\omega'\tau'}d\tau d\tau'\\ &=~ \int e^{-i(\omega - \omega')\tau}d\tau, \end{split} \end{equation} after integrating in $\tau'$. This integral does not converge, but in the sense of distributions, \begin{equation} \int e^{-i(\omega - \omega')\tau}d\tau = 2\pi\delta(\omega - \omega'). \end{equation}
What to do about the cross-terms? We must consider the integrals in $\omega$ and $\omega'$ separately for those. Consider the one for $\omega$: \begin{equation} \int\frac{\left<\widehat{y}(\omega)\right>}{i\omega - a}e^{i t\omega}d\omega \end{equation} If we assume analyticity of $\left<\widehat{y}(\omega)\right>$ (as a function of complex-valued $\omega$) and assume that it decays rapidly enough for large $|\omega|$ in the complex-$\omega$ plane, then we can appeal to contour integration and Cauchy's integral formula. The only pole in the complex-$\omega$ plane is found at $\omega = -ia$, where I assume $a$ is real. The integral is \begin{equation} \int\frac{\left<\widehat{y}(\omega)\right>}{i\omega - a}e^{i t\omega}d\omega ~=~ \lim_{N\to\infty}\oint_{\gamma_N}\frac{\left<\widehat{y}(\omega)\right>}{i\omega - a}e^{i t\omega}d\omega ~=~ 2\pi i\left<\widehat{y}(-ia)\right>e^{at}, \end{equation} where $\gamma_N$ is a path that includes the real interval $[-N,N]$ and a semi-circular arc connecting $N$ and $-N$. You can find examples of this path in almost any undergraduate complex analysis book. We assume that the integral on this arc decays to zero as $N\to\infty$.
I think this is how we pick up the exponentially decaying growing behavior in $\left<|x(t)|^2\right>$.
| improve this answer | |

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.