(180b) Understanding the Complex Reaction Network of Model Polyolefin Upcycling Via Informatics and Machine Learning

Searching for economical treatments of plastic waste is a prevailing topic in environmental issues. In contrast to traditional methods such as recycling, landfill or fuel, plastic upcycling - transforming plastic monomers into valuable products - appears to be a promising alternative. Despite the merits of upcycling, the inherent complexity of the reaction network made the experimental process of finding potential valuable molecules and their reaction pathways from waste plastic very time-consuming, considering the fact that thousands of husbands of molecules and reaction routes are involved.

In this talk, we present a computational workflow to speed up the screening of the complex reaction network to study the product distribution and synthesis routes in polyolefin upcycling. Using Decane (C10) as a representative polyolefin, the workflow: (1) builds the reaction network for a multifunctional metal-acid catalytic system using Rule Input Network Generator (or RING), (2) computes the minimum energy structure and vibrational properties of the intermediates of the network using the publicly available machine learning program, TorchANI, (3) evaluates the free energy of these intermediates using the harmonic approximation and statistical mechanics for ideal gas, (4) extracts likely pathways to various valuable products (aromatics for instance) via network traversal algorithms and thermodynamics-based screening. This procedure successfully revealed more than fifty aromatic compounds as potential upcycling products, many of which are experimentally reported for polyethylene upcycling. The thermodynamically favorable pathways and bottleneck reactions were also found. The extracted information can be compiled to map out the favorable pathways to different products such as substituted benzenes and polyaromatics.