ICAMS / Interdisciplinary Centre for Advanced Materials Simulation


Shooting point ensembles for likelihood maximization analysis

Date: 23.02.2009
Place: Rare Events in High-Dimensional Systems, Los Angeles (CA), USA

Jutta Rogal

The identification of a proper reaction coordinate is often a daunting task. In complex systems the reaction coordinate is usually expressed as a function of some collective variables that condense the full phase space into physically important variables. The choice of these collective variables though is not straightforward. Peters and Trout [1] have developed a likelihood maximization method to analyse a shooting point ensemble obtained in a transition path sampling (TPS) simulation to quantitatively identify the best reaction coordinate described by the most important collective variables. They use an aimless shooting algorithm in the TPS which is designed to keep the shooting points close to the transition state (TS) area. Here we discuss if their likelihood maximization analysis can also be used for other shooting point ensembles. In particular we focus on the one-way shooting algorithm. We find that we cannot use the entire shooting point ensemble if we want to approximate the reaction coordinate by a linear combination of collective variables since this linear approximation might only be true close to the TS. But even a non-linear description of the reaction coordinate is problematic since the value of the committor probability along the reaction coordinate changes significantly close to the TS and is insignificant in the tails of the function. A proper restriction of the shooting point ensemble is likewise not straightforward. Here we discuss some ideas how the shooting point ensemble from a one-way shooting TPS simulation might be suitably adjusted.

[1] B. Peters and B.L. Trout, J. Chem. Phys. 125, 054108 (2006)

Supporting information:

« back