Motivation

The motivation for ROLLO is to provide a user-friendly tool that helps reactor designers use evolutionary algorithms for reactor design optimization. Evolutionary algorithm setup is highly customizable with an assortment of genetic algorithm designs and operators. A reactor designer unfamiliar with evolutionary algorithms will have to go through the cumbersome process of customizing a genetic algorithm for their needs and determine which operators and hyperparameters work best for their problem. Furthermore, computing fitness values with nuclear software is computationally expensive, necessitating using supercomputers. Reactor designers have to set up parallelization to use the genetic algorithm optimization with nuclear software.

ROLLO provides a general genetic algorithm framework, sets up parallelization for the user, and promotes usability with an input file that only exposes mandatory parameters. ROLLO strives to be effective, flexible, open-source, parallel, reproducible, and usable:

• Effective: ROLLO is well documented, tested, and version-controlled on Github.

• Flexible: ROLLO is nuclear software agnostic. Users can vary any imaginable parameter with any nuclear software. ROLLO uses a templating method to edit the input file of the coupled software.

• Parallel: Users have two options for running ROLLO in parallel.

• Reproducible: Data from every ROLLO run saves into a unique pickled file.