Determining orbital localization from chain conformation in amorphous Polyethylene, using supervised learning methods

Hole transport in polyethylene (PE) can be modelled at the single electron level as a hopping process between occupied molecular orbitals (MO), these orbitals form states which extend along the chain backbone but are confined by torsions to shorter chain segments. If the chain is approximated as a discrete series of charge localized segments Marcus theory can be used to calculate the transfer rates between segments in the classical limit, and hole hopping transport can be simulated using kinetic Mote Caro (KMC). We have previously derived approximate segmentation rules based on sequences of trans and gauche torsions, and used these for KMC simulations of holes in amorphous PE. In this work we investigate the statistical correlations between chain conformation and MO localization in amorphous PE, with system coordinates generated using molecular dynamics and MOs obtained from density functional theory calculations in the CP2K software. Orbital localization is measured in terms of the inverse participation ratio and the chain structure is categorized in terms of dihedral angles and energies. Supervised learning methods are implemented to classify the MO localization length based on the local chain structure, and a model is developed to optimally segment PE chains in the amorphous phase.