(74a) Programming Colloidal Assembly into Aggregates and Crystals By Landscape Engineering

Colloidal particles possessing anisotropic interactions ("patchy colloids") represent a flexible building block for the self-assembly of a diversity of structural and functional materials, including stimuli responsive chains, mesoscale "molecules", encapsulants, and open crystal lattices with omnidirectional band gaps. Combining tools from statistical mechanics, molecular simulation, nonlinear manifold learning, and genetic algorithms, we have established an approach to reverse engineer patchy colloids to spontaneously self-assemble aggregates with desired structure and function.

We achieve this goal through a design paradigm that we term "landscape engineering" wherein we rationally sculpt the surface of the free energy landscape governing multibody assembly to favor the assembly of desired aggregates. First, we conduct nonlinear manifold learning to identify good low-dimensional collective variables parameterizing the many-body assembly, and conduct enhanced sampling directly in these collective variables. Second, we construct free energy surfaces in these collective variables and define the "fitness" of a particular building block design according to an objective function based on the topography of the landscape reflecting the thermodynamic stability and kinetic accessibility of the desired aggregate. Third, we employ state-of-the-art genetic algorithms to optimize the topography of the landscape by refining the colloidal building block architecture and chemistry.

We demonstrate our landscape engineering 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 open crystal lattices with omnidirectional photonic band gaps. In applications to icosahedral polyhedra, we design anisotropic interactions that lead to 76% improvements in aggregation rate over expert human designs, and in an application to a pyrochlore lattice design a colloid capable of hierarchical assembly first into tetrahedral building blocks and subsequent condensation into the stable crystal. Our approach integrates machine learning with domain expertise to provide a new and generally extensible tool for the inverse design of self-assembling materials.