Supervisors:

Livia Bartok-Partay (Chemistry), David Quigley (Physics), James Kermode (Engineering)

Summary:

Soft-matter systems exhibit an extremely rich macroscopic behaviour, with complex and fascinating phase properties. Many of these exotic structures and transitions can be captured by relatively simple hard-core soft-shell interaction potentials in simulations, however, our precise understanding is often obscured by difficulties in sampling the configuration space exhaustively. Linking such potentials to specific macroscopic properties requires many expensive simulations, limiting insight into how the form of the potential determines phase behaviour. In this project we will adapt a novel data-intensive sampling algorithm to make automated predictions of structural and thermodynamic properties. We will then exploit the new algorithm to reverse engineer models, using machine learning methods, that capture novel phase behaviour of specific soft-matter systems by design rather than discovery through brute-force trial and error.

Background:

Self-assembly of particles is often controlled by a competition between different interactions in soft matter systems, such as anisotropy, multiple characteristic length scales or the shape of the particles. These can induce the formation of a wide variety of structures and exotic phases, including open structures, mesophases, liquid polymorphism and quasi crystals. Computational modelling plays an important role both in understanding the microscopic properties of such soft materials and also in designing new materials with certain target properties, with practical application in colloid and polymer science, dust grains in plasma environments as well as photonic crystals. These systems are often represented in simulations by relatively simple model potentials, allowing tuneable properties and efficient large-scale calculations. However, unravelling the phase behaviour of these models is extremely challenging: we cannot rely on chemical intuition to predict complex and exotic phases, thermal and entropic effects make the use of global optimisation techniques limited, while observing the nucleation event and formation of thermodynamically stable solid phases in traditional simulations is equally problematic. As a result, our knowledge and understanding of the behaviour of these model systems is often biased, inconsistent or even misleading, hindering the ability to make practically useful predictions.

Research Questions and Aims of the PhD project:

• What is the real phase behaviour of widely used soft-matter model potentials ?
• How do charges and external fields affect these phase properties?
• What is the most efficient protocol to tune potential parameters, in order to “dial in” specific phase transitions?

In this project you will first examine a selected group of existing soft matter potentials and explore their phase behaviour using unbiased sampling techniques (such as nested sampling) to explore the configuration space exhaustively. This is expected to uncover previously unsampled parts of the parameter space, finding novel structures. In the second stage you will exploit these findings and tackle the inverse problem: i.e., design a protocol to reverse-engineer and parameterise potential functions behaving in a specific way, designed to reproduce desired phases, using machine-learning interatomic potentials frameworks.

Skills that the student will acquire: 

• Using configuration sampling techniques (nested sampling, Monte Carlo simulation) and evaluating thermodynamic properties
• Using machine learning interatomic potential frameworks
• Software development (python)
• Solving inverse problems with probabilistic methods

For further details about the project and how it links to the training included in the HetSys PhD programme, please visit click here