It doesn't work, though. On firefox 19.0.2 (archlinux), it seems to detect clicks shifted a good bit to the right (maybe 100 pixels?), and a little bit down. It looks like it's using whole-screen coordinates without correcting for the position of the canvas.

It would also be cool if you could give some quantitative data below - the total amount of kinetic/potential energy would be fun to watch, for instance. (And graph it.)

You also seem to be using an inverse-square law, which is "incorrect" in a 2D universe, but I suppose it makes it seem more natural.