5
$\begingroup$

I am trying to understand the connection between white noise and the Wiener process in the context of SDEs. At the beginning one starts with a differential equation including white noise $\xi_t$, e.g.,

$\frac{dX_t}{dt}=a(X_t,t)+b(X_t,t)\xi_t$.

and some initial value $X_0$. This is equivalent to the integral equation

$X_t=X_0+\int_{0}^{t}a(X_s,s)ds+\int_{0}^{t}b(X_s,s)\xi_sds$.

And now comes the step I do not understand. The second integral is equal to the Ito integral

$\int_{0}^{t}b(X_s,s)dW_s$

where $\{W_t\}_{t\geq0}$ is a Wiener process.

Why is that? How can one explain this in a mathematical rigorous way? Many textbooks just argue that the white noise is some kind of derivative of the Wiener process, i.e., $"\frac{dW}{dt}=\xi_t"$ but do not go into more detail. Why can we "replace $\xi_tdt$ by $dW_t$"?

For reference, here is the definition of a white noise process I am working with:

A white noise process is defined to be a generalized wide-sense stationary Gaussian process $Z_t$ with mean zero and covariance function $E[Z_sZ_t]=\delta_0(t-s)$. Here $\delta_0$ is the Dirac Delta function at 0.

$\endgroup$
3
  • 1
    $\begingroup$ What is a stochastic variable with variance infinity? How do you calculate with it? How can you insert such an un-variable into your first equation? $\endgroup$ Commented Jul 16, 2017 at 6:56
  • 2
    $\begingroup$ The $\xi$ presentation is not rigorous, the $W$ presentation is. $\endgroup$
    – Did
    Commented Jul 16, 2017 at 19:56
  • 1
    $\begingroup$ Thanks, this pointed me in the right direction. I will read some additional literature about this topic. Unfortunately this is often only very briefly discussed in introductory literature about sde. $\endgroup$
    – Spike
    Commented Jul 24, 2017 at 15:50

1 Answer 1

4
$\begingroup$

As I remember, a stationary process is a white noise if

  • it is a generalized wide-sense stationary process
  • it's spectral density is constant: ${s_\xi}_t(\nu)=\frac{1}{2\pi}\int_{-\infty}^{+\infty} e^{-is\nu}{K_\xi}_t(s)ds= c.$

One should remember the property of a white noise. If the covariance function equals to $${K_\xi}_t (\tau)=2\pi c \delta(\tau)$$ then $\xi_t$ is a white noise. In order to prove it one should use the definition given above and remember the property of the Dirac delta function: $$\int_{-\infty}^{+\infty}\ \delta(s) e^{-i\nu s }ds =1. $$ Now let's consider the derivative of the Wiener process $W_t$. It's covariance function equals to ${K_W}_t(t_1,t_2)=\sigma ^2 min(t_1,t_2)$. It's easy to see that the usual derivative of the Wiener process doesn't exist.
$$\frac{\partial {K_W}_t}{\partial t_1}=\left\lbrace \begin{matrix} \sigma^2,\, t_1<t_2\\ 0, \,\,t1>t_2 &\\ \end{matrix}\right. $$ But the distribution exists $$\frac{\partial^2 {K_W}_t}{\partial t_1 \partial t_2}= \sigma^2 \delta (t_2-t_1). $$ It follows from $E[W_t]=0$, that $E\left[\frac{d W_t}{dt}=0\right]$. The derivative of the Wiener process is also a wide-sense stationary process. We conclude that the derivative of the Wiener process is a white noise having intensity equals to $c=\frac{\sigma^2}{2\pi}$.

$\endgroup$

You must log in to answer this question.