|
Neuroevolutionary learning of particles and protocols for self-assembly
S. Whitelam, I. Tamblyn Physical Review Letters, 127, 018003 (2021) Designing particles and protocols that reliably drive self-assembly into desired structures remains a fundamental challenge in materials science and soft matter physics. In this work, we demonstrate that neuroevolutionary learning -- a technique that combines neural networks with evolutionary algorithms -- can design both particle interactions and time-dependent assembly protocols without relying on traditional assumptions about thermal equilibrium or mechanical stability. Using computational simulations of molecules on surfaces, the approach operates in two complementary modes: directed design, which creates materials with specific user-defined properties, and exploratory search, which discovers novel structures within specified order parameters by focusing on kinetically accessible configurations rather than just energetically favorable ones. This work opens the door to discovering assembly pathways and particle designs that would be difficult or impossible to find using conventional physics-based approaches alone. |
