PPARC logoFrontiers UK particle physics, astronomy and space science

select issue
Issue 1 Issue 2 Issue 3 Issue 4 Issue 5
Issue 6 Issue 7 Issue 8 Issue 9 Issue 10
Issue 11 Issue 12 Issue 13 Issue 14 Issue 15
Issue 16 Issue 17 Issue 18 Issue 19 Issue 20
Issue 21 Issue 22 Issue 23 Issue 24 Issue 25    
Full article list
 
 Update

 

go to top

back to issue 20 contents

RAL makes work for idle PCs

An AI program has been used to 'evolve' the perfect design for a neutrino factory by accessing the number-crunching power of home computers via the Internet

A Muon1 simulation
A Muon1 simulation, showing the losses associated with an overly-narrow solenoid (centre) and the remaining pion beam (right)
Work is underway at the Rutherford Appleton Laboratory to design a particle accelerator complex known as the Neutrino Factory (Frontiers 11). Its goal is to produce a downward-directed beam of neutrinos sufficiently intense to be detected on the other side of the Earth, after travelling thousands of kilometres through solid rock. This will allow physicists to calculate their mass.

The neutrino beam is produced by firing a high-power (4 megawatt) proton beam at a target material to generate large quantities of pions. The pions decay into muons which then produce the neutrinos when they, in turn, decay. One of the most challenging parts of the process is to capture the particles as they come off the target. This is difficult because the pions spray in all directions and must be focused by large superconducting solenoid magnets in order to form as intense a muon beam as possible. The situation is further complicated because particles also change direction as they decay into muons. How many are retained in the muon beam clearly affects the final neutrino intensity.

The Muon1 project

An important way to evaluate the best design is to simulate on a computer what happens to the particles in such a system. Computer codes are often used to calculate the efficiencies of accelerators, but we have taken the idea of computer-aided design a stage further, using artificial intelligence to design the accelerator system itself.

A 2D visualisation of the design space for an optimisation
A 2D visualisation of the design space for an optimisation
The Muon1 project uses the spare CPU time of several hundred volunteers' PCs to simulate possible accelerator beamlines while their computers are idle or being used for a less-intensive task such as word processing. The program finds better designs by a process of evolution: initially it simulates a soup of random designs, mostly with very poor efficiencies. The results are recorded and the better designs used for 'breeding' new ones by processes including mutation – where some parameters are slightly altered, and genetic crossover – where settings from two good designs are mixed. Periodically, the program returns the results over the Internet and can also download good designs generated by others.

The results from these optimisations often significantly surpass 'human designed' versions of the same accelerator and can potentially evaluate design options far more thoroughly – testing millions of configurations for each. The number of volunteers who downloaded the program was also a surprise: so far, we have seen almost 2000 different user-names in the results received, suggesting that public interest in cutting-edge physics is considerable – when they feel they can participate.

Stephen Brooks
E-mail: s.j.brooks@rl.ac.uk