(704b) Programmed Assembly of Anisotropic Patchy Colloids By Nonlinear Learning and Landscape Engineering
In this work we build on these foundations to perform inverse design of colloidal building blocks to spontaneously self-assemble into aggregates with desired structure. We engage this challenge through the principle of "landscape engineering" â rational sculpting of the low-dimensional assembly landscape to maximally favor the target morphology. Implementing this concept in practice, we couple our nonlinear machine learning approach with state-of-the-art optimization engines to iteratively optimize building block properties to maximize the thermodynamic stability and kinetic accessibility of the target structure. We demonstrate our approach in the inverse design of anisotropic patchy colloids engineered to assemble hollow Platonic polyhedra with applications as biomimetic models of viral capsids and in the hierarchical assembly of 3D photonic crystals. In an application to icosahedral target aggregates, our approach discovers a colloid architecture that produces ~45% improved yields than that designed by expert knowledge and geometric concerns alone. Starting from this optimized icosahedral design and selecting octahedra as our target, our approach is capable of efficiently mutating the building block design from this poor starting guess to engineer a new colloid programmed to assemble into octahedra with high yield and selectivity.