Hyper-parameter exploration of patchy particle design for self-limiting assembly

Inverse design of building block features for targeted self-assembly structures is now possible with recent advancements in machine learning and data-driven methods. However, due to the high dimensionality of the building block's parameter space, the final designed building block highly depends on the choice of the initial set of particle parameters to optimize. Sometimes, the optimized building block design can be completely different with a small change in the initial parameter set selection. In this work, we address the challenges of selecting appropriate particle parameters to optimize for self-limiting assembly using JAX-MD, an end-to-end differentiable molecular dynamics engine. We demonstrate that by choosing different sets of tune-able parameters to optimize in our patchy particle model, we not only achieve drastically different optimal building block designs for a given assembly target, but also affect the optimization success rate, without changing how the chosen parameters are optimized. These results will aid us in developing policies that promote further exploration into the design space when optimization fails.