Local Feature Analysis based Clustering Algorithm with Application to Polymer Model Reduction

We are interested in model reduction for the dynamics of large scale systems that contain a collection of spatially oriented points, in particular, a polymer system. Local feature analysis (LFA) introduces a specific state reduction algorithm that offers a topographic representation. In this paper, we propose a new LFA based clustering algorithm for system model reduction with application to the polymer system. The contribution is two-fold. First, a new sparsification algorithm is developed for LFA and offers a simple soft clustering rule. Compared to the existing empirical sparsification algorithm, the proposed method provides theoretical foundation. Second, we apply the proposed algorithm to a bulk glass polymer system in order to group atoms into superatoms, which is a critical step for polymer system model reduction. Several simulations are carried out to illustrate the application of the developed algorithm to the polymer dynamics.