Statistical identification of structural rearrangement events in molecular dynamics trajectories

We describe a statistical analysis method to detect simultaneous break points of the mean in multiple observables with Gaussian fluctuations as often observed in single molecule experiments and molecular dynamics trajectories. The maximum likelihood method finds the most likely break point in a given sequence of normally distributed multivariate observations using Hotteling's T2 statistic. In stochastic simulations we determine the threshold F-values for the acceptance of a break point in three dimensions and test the sensitivity and accuracy of the method. We present an algorithm that systematically identifies the most likely break points in a trajectory in a self-consistent manner. We test the method and algorithm with trajectories from molecular dynamics simulations on a prototypical atomistic glass former.