Skip to main content

Where Does cos(B) Spend Its Time?

Modified

November 12, 2026

The Question

The process \cos(B_{t}) starts at 1 and stays in [-1, 1]. Its mean is the real part of the characteristic function of B_{t} \sim \mathcal{N}(0, t) (Class 3), \operatorname{E}(e^{i B_{t}}) = e^{-t/2}, \qquad \text{so} \qquad \operatorname{E}(\cos B_{t}) = e^{-t/2} \to 0. \tag{1} Itô’s formula (Class 8) gives the same answer: M_{t} = e^{t/2} \cos B_{t} has \mathop{}\!\mathrm{d}M_{t} = -e^{t/2} \sin B_{t} \mathop{}\!\mathrm{d}B_{t}, so M is a martingale and \operatorname{E}(\cos B_{t}) = e^{-t/2} \operatorname{E}(M_{t}) = e^{-t/2}. Does a mean of zero mean that \cos B_{t} settles near zero? We simulate 5,000 paths and find the opposite.

Line chart over times 0 to 8. Five thin paths of the cosine of Brownian motion wander between minus 1 and 1 without shrinking. The simulated average and the exact mean, exp(-t/2), start at 1 and decay smoothly toward 0, lying on top of each other.
Figure 1: Five simulated paths of \cos B_{t}, the average over 5,000 paths, and the exact mean e^{-t/2}.

Where the Cosine Spends Its Time

Think of e^{i B_{t}} as a point on the unit circle with angle B_{t}. As t grows, the angle modulo 2\pi becomes uniform, and if \Theta is uniform then \operatorname{P}(\cos \Theta \leq x) = 1 - \arccos(x) / \pi, the arcsine law with density g(x) = \frac{1}{\pi \sqrt{1 - x^{2}}}, \qquad -1 < x < 1. \tag{2} The density is lowest at 0 and blows up at \pm 1: the cosine is flat at the top and bottom of the circle and steep at the sides, so it lingers near \pm 1 and rushes through 0. The circle is compact, so the same law also describes the fraction of time a single path spends in each region.

Two histogram panels sharing a density axis from 0 to 3. Left: at t = 0.25 the mass is piled up against 1; at t = 1 it is spread out but tilted toward 1; at t = 8 it is U-shaped and symmetric, on top of the dashed arcsine curve, which is lowest at 0 and rises sharply at minus 1 and 1. Right: the time-spent histogram of one long path is U-shaped and matches the arcsine curve.
Figure 2: Left: distribution of \cos B_{t} across 5,000 paths at three dates. Right: fraction of time one path spends in each bin over 0 \leq s \leq 1{,}000. The dashed line is the arcsine density (2).

Along that path, |\cos B_{s}| > 0.9 for 27% of the time and |\cos B_{s}| < 0.1 for only 7%, two sets of the same length; the arcsine law gives 29% and 6%. The mean is zero because the visits to +1 and -1 cancel, not because the cosine sits near 0.

The Path on the Cylinder

Plotting e^{i B_{t}} against time gives the curve (t, \cos B_{t}, \sin B_{t}) on a cylinder of radius one (Figure 3). The path winds back and forth at random and never leaves the surface, since |e^{i B_{t}}| = 1. The mean e^{-t/2} from (1) heads straight for the axis, because the paths spread evenly around the circle.

Three-dimensional plot with time on the long axis and the cosine and sine of Brownian motion on the other two. The path wraps irregularly around a faint unit cylinder, sometimes going forward and sometimes backward around it. A dashed line starts at the same point as the path and moves straight inward toward the time axis as time increases.
Figure 3: One path of (t, \cos B_{t}, \sin B_{t}) on the unit cylinder and the mean e^{-t/2} (dashed).