(622h) Detecting Transition Boundaries in Molecular Simulations

In computational studies of self-assembly, protein folding, and other molecular phenomenon, a key point of interest is the transition from one state to another. The detection of such transitions within a simulation are based on order parameters tuned to the particular phenomenon, but when the possible transition states are not known, an order parameter approach is difficult to use. Furthermore, defining the beginning of a smooth transition through a threshold as is often done is sub-optimal as either it fails to capture the beginning of an event or is too sensitive to random fluctuations within the system. We present in this talk an algorithmic method for determine the number, extent, and details of a transition using change point detection. Likewise, we will showcase multiple example phase transitions highlighting the performance and limitations of our method. The systems we explore will vary from hard particle to soft colloidal to atomic systems. We also will expound on potential applications of this technique in applications such as machine learning and active learning. By partitioning the system, machine learning can preferentially learn on transition states avoiding potential over-fitting on the long periods of stability common in simulations. We will also show how this can be used as a more traditional tool for computational scientist in exploring new systems.