(11c) Relentless: Development of a Python Package for Transparent, Reproducible Simulation-Based Optimization of Materials

The ability to design a material with specific properties is one of the biggest goals of materials engineering, with potential to make breakthroughs in catalysis, separations, sensing, energy storage, and other important engineering areas. Computational inverse design is a promising general approach because it does not have the costs of laboratory experiments and can quickly explore design candidates. Many inverse design approaches incorporate molecular dynamics simulations as a component of a larger optimization workflow using proprietary code or “spaghetti” scripts that limit transparency and reproducibility. In this work, we present relentless, an open-source Python package that automates the optimization of objective functions computed with simulations. relentless carries out simulations using existing molecular dynamics software (HOOMD-blue, LAMMPS) to take full advantage of their optimized performance. A model, simulation protocol, and optimization strategy are specified through a high-level, object-oriented Python interface — making relentless easy to use, to extend, and to connect with other scientific Python packages.