Modeling polymer meta-heuristic algorithms with thermal and electrical performance criteria

Here, we use a nature-mimicking optimization method, the genetic algorithm, alongside ML-based predictive models to design polymers whose practically useful but severe property criteria (i.e., glass transition temperature, Tg > ۵۰۰ K, and band gap, to eg > ۶ eV). Similar to nature, the characteristic properties of a polymer are assumed by the constituent types and the sequence of chemical building blocks (or fragments) in the monomer unit. Data-driven or machine learning (ML) methods have recently been used in materials science to provide rapid predictions of material properties. Although these predictive models are powerful and robust, they are still limited in their application to the design of materials with target properties or performance goals. The evolution of polymers by natural crossover, mutation and selection operations over ۱۰۰ generations resulted in ۱۳۲ new (compared to ۴ previously known) chemically unique polymers with high Tg and Eg. The chemical guidelines for the parts that make up the polymers with extreme thermal and electrical performance criteria are selected and revealed by the algorithm. The approach presented here is general and can be extended to design polymers with different characterization goals.

Keywords:

meta-heuristic algorithms , gray wolf algorithm , glass transition temperature , band gap