One of the best things about doing research is the ability to travel, either to conferences or for collaborations. I gave talks at a couple of conferences recently, so wrote up a brief descriptions of what I was talking about below:

European Fusion Physics Workshop 2012 (Ericeira, Portugal)

This is a series of workshops which focus on a particular issue in fusion. This time it was on "Power Exhaust in Fusion Devices", i.e. how to design and build a reactor which can handle the enormous amounts of power coming out of the plasma without melting (except if it's supposed to, like some liquid metal wall designs).

One of the crucial issues for fusion reactor design is handling the power coming out of the core plasma. In magnetic confinement fusion, strong magnetic fields are used to hold plasmas at around 100 million degrees in a toroidal (doughnut shaped) vessel. To keep the hot plasma core separated from the walls, this magnetic field is configured to have two regions: closed field-lines where the plasma core is confined, surrounded by open field-lines where it is not. The open field-lines channel plasma leaving the core onto armour-plated regions called "divertors", which are designed to handle the heat loads produced. Separating these two regions is a magnetic 'X-point'. This talk was about trying to understand how plasma behaves as it moves from the X-point to the divertor plate, and so improve predictions of the power onto divertor plates. This work involves improving our understanding plasma turbulence, the flow of heat and particles along complicated magnetic field-line geometries, and mixing of the plasma with neutral gas present in the vessel. One of the tools used for this is the BOUT++ 3D simulation code, which is developed at York with national and international collaborators, in particular LLNL.

8th International Conference on Computational Physics 2013 (Hong Kong)

Magnetised fusion plasmas are complex nonlinear systems, and are almost always turbulent. Simulating this is challenging, and requires supercomputers like HECToR with thousands of CPUs. As our simulation tools become more sophisticated, and the sizes of the machines we want to simulate become larger, making efficient use of these thousands of CPUs is increasingly important. The ICCP8 conference brings together researchers from many areas of computational physics to discuss techniques for efficient simulation of physical systems. The talk gave an outline of recent results from the BOUT++ code, and presented new work carried out at York in collaboration with Argonne National Laboratory (ANL) in the USA: A combination of mathematical analysis and numerical algorithms yielded an efficient "physics based" preconditioner, which resulted in a factor of 10 improvement in speed for some plasma simulation problems. This will allow the users of BOUT++ around the world to run more sophisticated simulations for longer, whilst using the same computational resources.

Paper: Non-local approach to kinetic effects on parallel transport in fluid models of the scrape-off layer

Post-doc John Omotani's first paper in plasma physics (he did his PhD in string theory). Published in PPCF here, but a pre-print can be found on arXiv here.

There are broadly two approaches to simulating plasmas: kinetic and fluid. Kinetic methods describe the particles in the plasma (ions and electrons) with a range of different energies, and can calculate how many of them have a particular energy (the "distribution"). This gives good results, but requires a lot of computing time in general. If the collisions between particles are frequent enough, then the distribution of particles follows an equation called the Maxwell-Boltzmann distribution. Fluid methods use this to reduce the number of calculations needed, at the cost of not being able to describe situations where the distribution is far from Maxwell-Boltzmann. In the edge of magnetic confinement fusion devices, like tokamaks, the collision rate is often high enough parallel to field-lines for ions to follow a Maxwell-Boltzmann distribution, but not electrons. This can result in the predictions of fluid models becoming quite inaccurate, particularly during transient events like ELMs. This is important because the power loads on material surfaces during ELMs depends in part on the heat conducted by electrons parallel to the field. This work is on ways to improve the fluid models by finding approximate solutions to the kinetic equations, without the computational expense of full kinetic calculations.