Soft Materials Design with Automatic Differentiation

Automatic differentiation (autodiff), the ability to compute gradients of arbitrary, complex functions, is the backbone of most machine learning and artificial intelligence algorithms. Instead of using autodiff implicitly in machine learning, we will use it explicitly by combining it with a molecular dynamics (MD) engine. By differentiating through molecular dynamics trajectories, we can directly inverse-design building-block properties for targeted materials assembly behavior. In this talk, I will introduce a framework for using an autodiff-enabled MD engine to optimize patchy particle parameters for targeted self-limiting assembly. I will highlight not only counterintuitive design results, but also demonstrate how we can use this framework to explore the interactions among different design parameters.