Machine learning free energy functionals for multiscale modelling

The broad scope of the project is to apply and develop neural network based classical density functional theory (cDFT) approaches to understand liquid mixtures and their interfaces with solid surfaces. The exact nature of the project can be tailored to suit the expertise and interest of the candidate, but possible areas of research include:

·     Developing cDFT to describe complex order parameters relevant to chemical systems (e.g., water, electrolytes, and ionic liquids).

·     Using cDFT and molecular simulations with accurate interatomic potentials to obtain an accurate thermodynamic description of liquid mixtures (e.g., phase diagrams and equations of state) relevant to industrial processes (e.g., extraction, distillation, and green recyclable solvents).

·     Incorporating an accurate description of polymeric systems (e.g., polyelectrolytes and soft matter systems) into the cDFT framework.

·      Developing and applying cDFT to understand response at the mesoscale in systems comprising liquid-solid interfaces, with a view to informing design of microdevices (e.g., for desalination and blue energy harvesting applications).

cDFT is a theoretical framework that permits efficient calculation of the structure of liquids and thermodynamic information. Although the principles underlying cDFT have been known for a long time, its application to complex chemical systems has been limited. Recently, it has been shown that neural networks can be used to greatly improve the accuracy and efficiency of cDFT, which provides the exciting opportunity to accurately describe complex chemical systems much more efficiently than with molecular simulations. Not only does this include vastly extending the length scales that can be studied, but it opens the possibility to developing cDFT approaches based on a first principles description of interatomic interactions. Recent work in the Cox Group has extended the use of this “neural cDFT” to mixtures and ionic solutions [see Bui and Cox, arXiv:2410.02556 (2024)].

In addition to learning fundamental theoretical concepts, this project will provide the candidate with the opportunity to develop their skills in programming and gain experience in machine learning techniques. Suitable candidates will have a strong background (such as a Masters-level degree with a research component) in computational/theoretical chemistry or related discipline. A background in statistical mechanics is highly desirable.

If you are interested, please contact Steve Cox (stephen.j.cox@durham.ac.uk) with a copy of your CV and a brief description of any previous research experience.