PhD project: Machine Learning interatomic forcefields for shockwaves and phase transitions

Project description

Molecular dynamics simulation is the most widely used technique in materials modelling. It relies entirely on the model describing  forces acting between atoms. 

The traditional approach was to make a model for forces containing as much physics as possible, identify the main contributions to binding, cut out the computationally expensive parts and fit parameters to material properties.   Nowadays, fully quantum mechanical calculations enable us to generate a huge database which can be using to train these models.   While the language has shifted from  "fitting a model" to "training a neural net" the essence of the problem is still to have an efficient method of converting a set of atomic positions into a set of atomic forces.

In this project you will use physical insight to generate the potential, which may be a simple analytic function or a full-blown neural net.  Then you will generate the relevant data needed to determine the model parameters (training).  Finally, you will use the model to determine materials properties under extreme conditions which would be impossible to study in detail experimentally.

Typical applications will include shock waves passing through materials.  Shock is known to raise the pressure and temperature of materials to astronomical levels, but early studies using Free Electron Lasers are suggesting that the state of these materials depends not just of thermodynamics, but also on nonequilibrium behaviour during phase transitions.  In some cases entirely different structures are observed in shock from what is known to be the thermodynamic equilibrium.  There is currently no theoretical framework to determine what non-equilibrium phases will appear: this simulational work will tackle the problem.

Of particular interest to the group are light molecules - hydrogen, water, methane, ammonia - and metallic alloys and elements - sodium, potassium titanium. gallium, zirconium, bismuth, arsenic.

It is possible to study this project with advanced training in our MAC-Migs CDT - More information is here,

Project supervisors

The project supervisors welcome informal enquiries about this project.

Find out more about this research area

The links below summarise our research in the area(s) relevant to this project:

What next?

More PhD projects