The total increase occurs at a countable set of jumps. The simulation gives some idea of this.

Bugs:

- The vertical axis is continuously rescaled so that the portion of the process visible in the window fills the window. This is arguably wrong, because it makes the zooming look funny, but also arguably right, because it works for stochastic processes that are non-self-similar, like this one.
- Eventually the discreteness of computer floating-point arithmetic wrecks the simulation. When the total variation of the process in the display window is smaller than the smallest representable positive number, we get a flat line.

For some statistical Java applets by other people, see http://www.stat.duke.edu/sites/java.html.

Back to my home page.