Thinking backwards - major prize awarded to EPSRC Research Fellow for numerical software

Supplementary content information

Patrick Farrell

Dr Patrick Farrell, an EPSRC Early Career Research Fellow in the Mathematical Institute of the University of Oxford, has won the 2015 Wilkinson Prize for numerical software, which is awarded once every four years.

Patrick's research on adjoint derivation has produced a new mathematical technique which addresses a major problem in scientific computing.

Patrick was the lead author of the dolfin-adjoint software, which automatically derives and solves the adjoint equations for a very general kind of discretisation. Patrick's co-authors, David Ham, Marie Rognes and Simon Funke also shared the prize.

Patrick is currently applying the technique on two novel engineering projects as part of his EPSRC fellowship;

  • identifying the optimal placement for marine turbines to maximise power for renewable tidal energy - crucial for the success of this £76 billion industry
  • developing tools to better understand and diagnose ischaemic heart disease

The dolfin-adjoint software is built on top of the FEniCS project and is open-source. It is being used by other scientists in areas such as understanding the ice sheets on Greenland and the Antarctic, elastic imaging for identifying cancerous tissues, understanding subsurface geological features, and the optimal design of load-bearing structures.

What is the technique?

The adjoint technique allows an engineer to design backwards from a desired end goal, rather than forwards by trial and error, and has very important applications across the whole of computational science and engineering.

Patrick explains: We typically write down the laws of physics as partial differential equations. These equations propagate causality. You put in some causes such as the shape of an aeroplane, or the weather today, and the equations propagate effects - telling you how well the aeroplane flies or the weather tomorrow.

Now, associated with every such equation (the 'forward' equation) is associated another equation, the 'adjoint' equation, which propagates causality backwards: it maps in a linear way from effects back to their causes.

These adjoint equations are important because they tell you how to improve your design: instead of asking, 'what is the drag coefficient of this wing', the adjoint lets you ask, 'what is the wing shape that has the best drag coefficient?'

Patrick continues, It was an enormous surprise when I received the telephone call from the chairman of the Wilkinson prize committee - I hadn't expected to win at all when there is such an enormous amount of high-quality numerical software available.

I'm delighted and honoured that my work has been recognised in this way, and it would not have been possible without my EPSRC fellowship.

The Wilkinson prize will be awarded at the International Congress of Industrial and Applied Mathematics (ICIAM) in Beijing in August.

EPSRC's Fellowship scheme is open to scientists at all stages of their career (details of the areas in which fellowships are available). In particular, the areas of 'Software development for novel Engineering research' and 'Software development for novel Physical Sciences research' are highlighted. There are currently 298 EPSRC Fellows carrying out research of UK and global significance.