Intuitively it seems likely that the expected whereabouts of Brownian motion on the unit circle would be the origin $\left(0,0\right)$, at least in the limit as $t\to\infty$. Is this right? Are there any techniques for determining the expected location for finite $t$ given some starting point $p$?

  • $\begingroup$ Brownian motion lies on the circle $x^2+y^2=1$ for all time. $\endgroup$ – user339017 May 12 '16 at 9:15
  • $\begingroup$ You can show that the wrapped normal distribution will converge in distribution to a uniform distribution as $t$ increases making the mean approach the origin along the radius from $p$, though the mode will stay at $p$. You might be able to calculate intermediate values by messy integration or simulation $\endgroup$ – Henry May 12 '16 at 9:28
  • $\begingroup$ I'll have a look into this. Thanks. $\endgroup$ – user339017 May 12 '16 at 9:49

Suppose $p$ is $(1,0)$ - if it is not then simply rotate the answer - and that the underlying Brownian motion is a standard Wiener process wrapped around the unit circle

A standard Wiener process has $W_t$ normally distributed with mean $0$ and variance $t$, i.e. standard deviation $\sqrt{t}$. Wrap this round the unit circle and you get $X_t = \cos W_t$ and $Y_t= \sin W_t$. You then have an mean expected value given by $$\mathbb{E}[X_t] = \int_{-\infty}^{\infty} \cos (w) \frac{1}{\sqrt{2\pi t}}e^{-\frac{w^2}{2t}}\,dw $$ $$\mathbb{E}[Y_t] = \int_{-\infty}^{\infty} \sin (w) \frac{1}{\sqrt{2\pi t}}e^{-\frac{w^2}{2t}}\,dw $$

By symmetry this makes $\mathbb{E}[Y_t] =0$, which is not a surprise

Empirically, it seems $$\mathbb{E}[X_t]=e^{-t/2}$$ or at least very close to that expression. Attempting to do the integration numerically in R with

cases <- 100000
n <- ((1:cases)-1/2)/cases
g <- qnorm(n)
t <- (0:1000)/100
expectedx <- rep(NA,length(t))
for (j in 1:length(t)){
  expectedx[j] <- mean(cos(g*sqrt(t[j])))
plot(expectedx ~ t, type="l")

gives a graph

E X versus t

and the maximum difference from $e^{-t/2}$ is less than $10^{-5}$, which is possibly due to the numerical approximation to the integral. This may also be related to the von Mises distribution, a close approximation to the wrapped normal distribution. Both converge in distribution to a uniform distribution on the circle as the underlying dispersion increases, and $\mathbb{E}[X_t] \to 0$ as $t \to \infty$


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.