skip to primary navigationskip to content


Data-driven design of soft materials using machine learning

Supervisor: Dr Alpha Lee

Predicting the structure of soft materials given the microscopic interactions between the constituent components is a significant hurdle in materials discovery. The challenge lies in disentangling the many timescales and lengthscales of the system. Advances in analytical theories (e.g. liquid state integral equation theories, dynamic density functional theory) have provided invaluable physical frameworks to think about ways to construct multiscale models. However, deriving the link between microscopic chemistry and coarse grain parameters analytically is often impossible, and the literature often relies on ad hoc approaches.  


In this project, we will combine techniques in machine learning with physical frameworks derived from statistical physics to infer and understand multiscale models using molecular simulations data as input. The first stage of the project will focus on equilibrium properties. We will infer liquid state theory models using high throughput simulations, and uncovering possibly “hidden” structures in liquids and soft materials using technique in unsupervised learning. Those data-driven models will also be applied to tackle problems such as solvation free energy prediction and inverse liquid design. The second stage of the project will focus on non-equilibrium transport properties, such as automatically inferring continuum equations from molecular simulations.