This is why you shouldn't do homework two days before it's due

My molecular dynamics simulations are being contrary. As in "contrary to what is known about molecular dynamics".

Suppose you have a particle (or two, or n) and you set it loose under some classical set of forces. You do this by calculating the position, velocity, and acceleration of the particle(s) at regular intervals (timesteps) iteratively: use the previous set of information to calculate the next one (basically, think Taylor series; that's pretty much all it is). Things like average energy can be derived from all this.

I'm pretty sure that the following are true:
1. smaller timesteps = better simulation, in the sense that standard deviations tend to be smaller and you catch "fast" events, but it also means the simulation is slow in some sense
2. the longer the simulation runs, the less it represents reality. Slow events are hard to simulate because of this.

Of course, *my* simulations flew in the face of all this: the larger the time step (until divergence), the smaller the standard deviation. Long simulations (say 2-3 orders of magnitude longer) behave practically the same as short ones.

The worst part is, the code isn't mine. It's given as part of the assignment. Which means it well ought to work, dammit!

Grrr.