Speaker
Description
Biological molecular machines reliably undergo complex nanoscale dynamics to perform tasks, demonstrating how activity and multistability can be harnessed to process information and do work. While it is now possible to engineer complex nanostructures from synthetic building blocks, endowing them with machine-like functionality remains elusive, in part because we lack tools to control their kinetics and tune how they switch between stable states. To take full advantage of our engineering capabilities we need tools that connect the particle level design attributes to the emergent dynamic behavior of assembled systems.
I will show how combining path reweighting, a tool for extrapolating existing simulation data to a modified Hamiltonian, with automatic differentiation makes it possible to compute gradients of dynamical observables, such as transition rates between metastable states, without storing or differentiating through simulation trajectories. This enables the desired functionality of a system to be formulated as an optimization problem and solved using gradient-based methods, providing a principled route to designing systems that switch between target states with prescribed probabilities.
As a proof of concept, I will demonstrate this framework on passive as well as active particles navigating a rugged two-dimensional energy landscape with multiple metastable states. By optimizing the landscape, we can direct the system toward one or several target states with prescribed probabilities, illustrating how rational design principles can be used to encode life-like, adaptive behavior into multistable systems.